CN118312923B

CN118312923B - Intelligent park-oriented equipment measurement method and computer equipment

Info

Publication number: CN118312923B
Application number: CN202410721197.1A
Authority: CN
Inventors: 葛君
Original assignee: Zhuhai Yizhi Technology Co ltd
Current assignee: Ge Jun
Priority date: 2024-06-05
Filing date: 2024-06-05
Publication date: 2024-09-13
Anticipated expiration: 2044-06-05
Also published as: CN118312923A

Abstract

The application discloses an intelligent park-oriented device measurement method and computer equipment, wherein the method comprises the steps of obtaining original measurement data of devices in an intelligent park, carrying out data cleaning and format conversion, obtaining standardized target data for time sequence segmentation, extracting feature vectors of segmented data fragments, decomposing each data fragment, obtaining a plurality of component signals, calculating complexity indexes of each component signal to determine abnormal complexity indexes, obtaining fault information corresponding to the data fragments with corresponding abnormal component signals, and constructing a device fault ontology library according to the plurality of feature vectors and the fault information corresponding to each feature vector; extracting semantic features from real-time measurement data of the equipment to be measured, carrying out semantic matching and reasoning on the semantic features and the equipment fault ontology library, carrying out fusion analysis on sub-diagnosis results in the obtained primary diagnosis results, and determining a measurement method of the equipment to be measured according to the obtained final diagnosis conclusion and abnormal component signals.

Description

Intelligent park-oriented equipment measurement method and computer equipment

Technical Field

The application relates to the technical field of fault diagnosis, in particular to an intelligent park-oriented equipment measurement method and computer equipment.

Background

With the rapid development of industry 4.0 and intelligent manufacturing, the equipment operation and maintenance management of modern industrial parks is facing unprecedented challenges. On the one hand, the equipment in the park is various in variety and complex in working condition, the equipment is closely related with each other, the traditional post-maintenance and regular maintenance strategies are simply relied on, and the requirements of high-efficiency, reliable and long-period operation of the equipment are difficult to meet. On the other hand, in the whole life cycle of the equipment, all links from state monitoring, fault diagnosis and maintenance decision are often independent, lack of data fusion and flow penetration, and are difficult to realize global optimization and dynamic adaptation.

In recent years, state monitoring and fault diagnosis techniques have received extensive attention from industry and academia. The technology utilizes sensors to collect key parameters such as vibration, temperature, pressure and the like of equipment in real time, and utilizes methods such as signal processing, machine learning and the like to analyze monitoring data, so that the evaluation of the health state of the equipment and the identification of fault symptoms are realized. However, most existing state monitoring and fault diagnosis methods still have some drawbacks. First, in the data acquisition and processing stage, due to lack of deep understanding of the mechanism and failure mode of the device, selection of monitoring parameters and arrangement of measuring points often depend on experience and specifications, and it is difficult to ensure quality and comprehensiveness of data. Secondly, in the stage of abnormality detection and fault diagnosis, the common statistical control limit and expert rule method is easy to generate false alarm and missing alarm, and the diagnosis has poor interpretation. For complex systems and complex faults, their applicability and accuracy are limited.

In the aspect of maintenance decision optimization, the conventional method is to make a maintenance plan with a fixed period based on reliability theory and expert experience. Such static strategies do not adequately take into account the actual health and risk levels of the equipment, often resulting in wasted maintenance costs and reduced equipment availability. In recent years, some students have tried to apply the results of state monitoring and fault diagnosis to the optimization of maintenance decisions, and state-based maintenance (Condition-Based Maintenance, CBM) strategies have been proposed. However, most of these CBM methods are directed to a single facility or independent production line, and lack overall consideration for overall operation and resource scheduling of the campus facility. Furthermore, they are not fully considered for equipment degradation mechanisms and uncertainties in fault prediction and maintenance planning, and the dynamics and robustness of the optimization remain to be enhanced.

In general, how to realize the collaborative optimization of the whole life cycle management of equipment and the production operation of a park has become a key problem to be solved urgently in the construction of an intelligent park.

Disclosure of Invention

The embodiment of the application provides an intelligent park-oriented equipment measurement method and computer equipment, which can solve the problem that the current maintenance decision optimization method is mostly aimed at a single equipment or an independent production line and lacks overall consideration of overall operation and resource scheduling of park equipment. In addition, in the process of fault prediction and maintenance planning, the consideration of equipment degradation mechanism and uncertainty is insufficient, and the optimization dynamics and robustness are to be enhanced. The method can realize the collaborative optimization of the whole life cycle management of the equipment and the production operation of the park.

In a first aspect, an embodiment of the present application provides an apparatus measurement method for an intelligent park, where the method includes:

acquiring a plurality of original measurement data of preset equipment in an intelligent park, and performing data cleaning and format conversion on the original measurement data to acquire standardized target data;

Performing time sequence segmentation on the target data to obtain a plurality of data fragments, and extracting feature vectors of each data fragment; decomposing each data segment to obtain a plurality of component signals, calculating a complexity index corresponding to each component signal, and determining an abnormal complexity index in the plurality of complexity indexes;

acquiring fault information corresponding to the data segment according to the abnormal complexity index and an abnormal component signal corresponding to the abnormal complexity index, and constructing an equipment fault ontology library according to a plurality of feature vectors and fault information corresponding to each feature vector;

Acquiring real-time measurement data of equipment to be measured, extracting semantic features from the real-time measurement data, carrying out semantic matching and reasoning on the semantic features and the equipment fault ontology library, and acquiring a preliminary diagnosis result corresponding to the real-time measurement data; and carrying out fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, and generating a measurement strategy according to the final diagnosis conclusion and the abnormal component signal so as to determine a measurement method of the equipment to be measured.

In a second aspect, the present application also provides an apparatus for measuring a device for an intelligent park, including:

the data acquisition module is used for acquiring a plurality of original measurement data of preset equipment in the intelligent park, and carrying out data cleaning and format conversion on the original measurement data to acquire standardized target data;

The time sequence segmentation module is used for performing time sequence segmentation on the target data, acquiring a plurality of data fragments and extracting the feature vector of each data fragment;

The complex computing module is used for decomposing each data segment to obtain a plurality of component signals, computing a complexity index corresponding to each component signal and determining an abnormal complexity index from the plurality of complexity indexes;

the fault construction module is used for acquiring fault information corresponding to the data segment according to the abnormal complexity index and the abnormal component signal corresponding to the abnormal complexity index, and constructing an equipment fault ontology library according to a plurality of feature vectors and the fault information corresponding to each feature vector;

the primary diagnosis module is used for acquiring real-time measurement data of the equipment to be measured, extracting semantic features from the real-time measurement data, carrying out semantic matching and reasoning on the semantic features and the equipment fault ontology library, and acquiring a primary diagnosis result corresponding to the real-time measurement data;

and the final diagnosis module is used for carrying out fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, and generating a measurement strategy according to the final diagnosis conclusion and the abnormal component signal so as to determine a measurement method of the equipment to be measured.

In a third aspect, the present application also provides a computer device comprising a processor and a memory for storing a computer program which, when executed by the processor, implements the smart park oriented device measurement method of the first aspect.

In a fourth aspect, the present application also provides a computer readable storage medium storing a computer program which when executed by a processor implements the smart park oriented device measurement method of the first aspect.

Compared with the prior art, the application has at least the following beneficial effects:

1. The accuracy and the reliability of equipment fault diagnosis are improved. The method provided by the application adopts a self-adaptive measurement strategy and multi-source heterogeneous data fusion analysis, fully excavates fault characteristics and rules in equipment operation data, and builds an intelligent diagnosis model covering multidimensional knowledge of equipment parts, fault modes, abnormal symptoms and the like. Through intelligent algorithms such as semantic reasoning, mathematical statistics, machine learning and the like, the scheme can automatically identify the degradation state and fault mode of the equipment from mass monitoring data, and perform deep verification and optimization on the primary diagnosis result, thereby ensuring the accuracy and reliability of fault diagnosis, reducing diagnosis errors such as false alarm, missing alarm and the like, and providing reliable decision basis for equipment operation and maintenance.

2. And the predictive maintenance and active prevention and control of equipment faults are realized. The method provided by the application creatively provides a predictive maintenance strategy integrating fault diagnosis, self-adaptive measurement and early warning threshold optimization. By tracking the dynamic change of the health state of the equipment in real time and adaptively adjusting the measurement parameters and the early warning threshold value, the scheme can acutely capture early degradation and tiny abnormality of the performance of the equipment, accurately estimate the residual service life and fault risk of the equipment, and provide powerful support for making a targeted overhaul plan and maintenance decision. Through advanced sensing and active prevention and control, the application can effectively reduce unexpected shutdown and performance degradation of equipment, prolong the service life of the equipment and improve the production efficiency and reliability of the equipment.

3. And the whole life cycle management of park equipment is optimized, and the comprehensive operation benefit is improved. The method provided by the application takes data driving and model enabling as cores, and carries out flow reconstruction and model integration on links such as equipment fault diagnosis, early warning, maintenance and the like, so as to form an intelligent management system for closed-loop optimization of the whole life cycle. Through multi-objective optimization solution of the equipment maintenance strategy, the scheme can reduce the detection maintenance cost and the resource consumption to the maximum extent while guaranteeing the reliability and the availability of the equipment, and can realize the maximization of benefits in the whole life cycle of the equipment. In addition, the method provided by the application carries out systematic knowledge extraction and intelligent modeling on the diagnosis rules and expert experience of equipment management, promotes the sediment accumulation and intelligent multiplexing of key diagnosis and maintenance technologies and experience knowledge, and is beneficial to the continuous improvement and innovation upgrading of the operation and maintenance management of a park.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application as claimed.

Drawings

FIG. 1 is a flow chart of a method for measuring equipment for an intelligent park according to an embodiment of the application;

FIG. 2 is a schematic diagram of a device measurement apparatus for smart park according to an embodiment of the present application;

fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present application.

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth such as the particular system architecture, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. It will be apparent, however, to one skilled in the art that the present application may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

It should be understood that the terms "comprises" and/or "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.

It should also be understood that the term "and/or" as used in the present specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.

As used in the present description and the appended claims, the term "if" may be interpreted in context as "when …" or "once" or "in response to a determination" or "in response to detection. Similarly, the phrase "if a determination" or "if a [ described condition or event ] is detected" may be interpreted in the context of meaning "upon determination" or "in response to determination" or "upon detection of a [ described condition or event ]" or "in response to detection of a [ described condition or event ]".

Furthermore, the terms "first," "second," "third," and the like in the description of the present specification and in the appended claims, are used for distinguishing between descriptions and not necessarily for indicating or implying a relative importance.

Reference in the specification to "one embodiment" or "some embodiments" or the like means that a particular feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the application. Thus, appearances of the phrases "in one embodiment," "in some embodiments," "in other embodiments," and the like in the specification are not necessarily all referring to the same embodiment, but mean "one or more but not all embodiments" unless expressly specified otherwise. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless expressly specified otherwise.

The following describes the technical scheme of the embodiment of the application.

Referring to fig. 1, fig. 1 is a flow chart of an apparatus measurement method for an intelligent park according to an embodiment of the application. The device measurement method for the intelligent park can be applied to computer devices including, but not limited to, smart phones, notebook computers, tablet computers, desktop computers, physical servers, cloud servers and the like. As shown in fig. 1, the smart park-oriented device measurement method of the present embodiment includes steps S101 to S106, which are described in detail below: step S101, a plurality of original measurement data of preset devices in the intelligent park are obtained, and data cleaning and format conversion are carried out on the original measurement data to obtain standardized target data.

Specifically, before determining the corresponding measurement methods for the devices in the smart park, raw measurement data in the working process of the devices in the smart park needs to be acquired, where the raw measurement data may come from multiple sources such as various sensors, meters, control systems, log files, and the like. The types of raw measurement data may include various physical quantities such as temperature, pressure, flow, vibration, current, voltage, power, etc., and unstructured data such as operating states of the device, alarm information, maintenance records, etc. In order to ensure the integrity and consistency of the original measured data, standard specifications and interface protocols for data acquisition need to be established, and requirements of data acquisition frequency, time stamp, data format and the like are defined. Meanwhile, security and privacy protection of data are fully considered, sensitive data are subjected to desensitization treatment, and a strict authority control mechanism is established.

After the raw measurement data is acquired, it needs to be preprocessed to improve the data quality and usability. The first step in the preprocessing is data cleansing, which aims to identify and process missing values, outliers, repeated values and inconsistencies in the raw measurement data. Common data cleaning methods include deletion, interpolation, average, median, and the like. For example, with respect to data collected by the temperature sensor, if the temperature value at a certain point in time is far beyond a normal range (e.g., 500 ℃), it may be recognized as an abnormal value and subjected to deletion or replacement processing. For another example, for data missing from a device during a certain period of time, the missing value may be estimated by interpolation (e.g., linear interpolation, spline interpolation), or the data for the period of time may be deleted directly. The process of data cleansing needs to rely on predefined rules and thresholds, which can be obtained based on expert knowledge, statistical analysis or machine learning methods.

After the data cleaning is completed, data conversion and integration are required. Raw measurement data from different sources may take different formats and units that need to be converted to a unified format and unit for subsequent analysis and processing. For example, for time series data, it is necessary to convert the time stamp into a standard date-time format (e.g., ISO 8601) and take into account factors such as time zone, daylight savings time, and the like. For different units of physical quantity, unit conversion is required, such as converting pressure from pascal (Pa) to megapascal (MPa). In addition, the accuracy and resolution of the processed data is required, such as keeping the floating point number to the last two bits of the decimal point, or downsampling the high frequency sampled data to a specified time interval. The data integration is to combine data from different sources into a unified data set, and the problems of mapping, synchronization, redundancy and the like of the data need to be solved.

In data conversion and integration, the interpretability and semantic interoperability of raw measurement data also need to be considered. Metadata and semantic annotations can be introduced to explicitly describe the meaning, source, quality and other attribute information of the data. Therefore, sharing and understanding of data among different systems and applications can be facilitated, and the multiplexing value of the data is improved. Semantic annotation can map data to standardized concepts and relationships by adopting knowledge organization systems such as ontology, word list and the like. For example, a device failure ontology may be defined that maps failure data from different sources onto a unified failure type and failure mode.

After data conversion and integration, data normalization and standardization processing are also needed. Normalization is the scaling of data to a specified range (e.g., [0,1] or [ -1,1 ]), which can eliminate the effects of different scales and orders of magnitude. Common normalization methods include min-max normalization, zero-mean normalization, etc. The normalization is to convert the data into a distribution with the mean value of 0 and the standard deviation of 1, and the comparison and analysis between different data can be conveniently carried out. The normalization and standardization processing can improve the comparability and stability of the data, and lay a foundation for subsequent analysis such as feature extraction, anomaly detection and the like.

And finally, carrying out quality verification and evaluation on target data corresponding to the preprocessed original measurement data, and ensuring that the target data meets the requirement of subsequent analysis. The quality index of the data such as the integrity rate, the accuracy rate, the consistency and the like can be calculated by adopting a statistical method and compared with a predefined quality threshold. For example, it is required that the integrity rate of the target data is not less than 99%, the error rate is not more than 1%, and the like. For target data which does not meet the quality requirement, correction or re-acquisition processing is required. By data acquisition and preprocessing, raw, heterogeneous, inconsistent measurement data can be converted into a standardized, high-quality target data set consisting of a plurality of target data. Step S102, time sequence segmentation is carried out on target data, a plurality of data fragments are obtained, and feature vectors of each data fragment are extracted. Specifically, for the target data after preprocessing, the application adopts a self-adaptive time window method to carry out time sequence segmentation. The traditional fixed time window method has certain limitation and cannot adapt to the dynamic change characteristic of the running state of equipment. In particular, the present application defines a local complexity index of a signal for measuring the non-linearity, non-stationary degree of the signal in a local time range. The index comprehensively considers the aspects of local fluctuation, mutation, trend and the like of the signal. When the local complexity of the signal is higher, the signal is shown to be changed drastically in a local time range, and the application adopts a smaller time window to capture the detail characteristics of the signal; when the local complexity of the signal is low, indicating that the signal is relatively stationary within a local time frame, the present application employs a large time window to capture the overall trend characteristics of the signal. By adaptively adjusting the size of the time window, the application can extract multi-scale characteristics of the running state of the equipment on different time scales, and key detailed information is not omitted, and noise and redundant information are not interfered.

For each segment of data within the adaptive time window, the present application extracts a set of statistical and shape features. The conventional feature extraction method usually only focuses on the statistical characteristics of the data, such as mean, variance, peak-to-peak value, and the like, and ignores the shape characteristics of the data. However, anomalies in the operating state of the equipment are often manifested as distortions and distortions of the waveform, such as pulses, spikes, truncated, distortions, and the like. To capture these shape features, the present application introduces novel form factors such as peak factor, pulse factor, kurtosis factor, skew factor, and the like. The peak factor is the ratio of the maximum absolute value to the root mean square of the data within a data segment, reflecting the peak level of the data. The pulse factor is the ratio of the maximum absolute value to the absolute average value of the data within a data segment, reflecting the intensity of the spike in the data. The kurtosis factor is the ratio of the square of the fourth-order central moment of the data in the data segment to the variance, reflecting the tail characteristic of the data distribution. The skew factor is the ratio of the third-order central moment of the data in the data segment to the cube of the standard deviation, reflecting the symmetry of the data distribution. By extracting the shape factors, the application can quantitatively characterize the waveform characteristics of the running state of the equipment and discover the abnormal symptoms of the equipment as soon as possible.

In some embodiments, the decomposing each of the data segments to obtain a plurality of component signals includes: acquiring energy distribution information, local extremum information and trip point information of a frequency spectrum corresponding to the data fragment; carrying out self-adaptive segmentation on the frequency spectrum corresponding to the data fragment according to the energy distribution information, the local extremum information and the trip point information to obtain a plurality of frequency bands; constructing a wavelet basis function corresponding to each frequency band; and decomposing the frequency band according to each wavelet basis function based on an empirical wavelet transformation method to obtain a plurality of component signals.

In the aspect of signal decomposition, the application adopts an Empirical Wavelet Transform (EWT) method to adaptively decompose the equipment operation state signal into a plurality of tightly-orthogonal frequency bands. The conventional wavelet transform method needs to preset wavelet basis functions, such as Daubechies wavelet and Morlet wavelet, and the selection of the basis functions usually depends on experience and trial and error, so that the characteristics of different types of signals are difficult to adapt. The EWT method can adaptively construct a wavelet basis function according to the frequency spectrum characteristics of the signals, so that the accuracy and the efficiency of signal decomposition can be improved, and the interference of human factors can be avoided. Specifically, the EWT adaptively segments the spectrum of the signal first, dividing the spectrum into a number of closely spaced frequency bands. The spectrum segmentation process comprehensively considers the characteristics of spectrum energy distribution, local extremum, jump points and the like, ensures the continuity and smoothness of the spectrum in each frequency band, and separates oscillation modes with different scales as far as possible. Then, a tightly orthogonal wave function is constructed in each frequency band as a wavelet basis function of the frequency band. The process of constructing the wavelet basis function adopts a Meyer wavelet generation method, and ensures the compactness, orthogonality and smoothness of the wave function. And finally, carrying out wavelet decomposition on the signals by using the wavelet basis functions with the self-adaptive structures to obtain a plurality of tightly-supporting orthogonal wave band signals. Each band signal represents an oscillation mode of the original signal on different frequency scales, and the multi-scale frequency domain characteristics of the running state of the equipment are reflected.

For each wave band signal obtained by EWT decomposition, the application also extracts the statistical characteristics and shape characteristics, such as mean value, root mean square, peak factor and the like, to form the characteristic vector of the wave band. Since the EWT decomposition is a recursive process, each stage of decomposition generates a plurality of sub-bands, and therefore, the application adopts a tree-like feature vector organization mode, which is called an EWT feature tree. The root node of the EWT feature tree represents the feature vector of the original signal, each intermediate node represents the feature vector of one sub-band, and the leaf nodes represent the feature vector of the final base band. The EWT feature tree not only contains feature information of different frequency scales, but also retains hierarchical and dependency relationships among the features. Through the EWT feature tree, the application can more comprehensively and systematically describe the frequency domain features of the running state of the equipment.

Illustratively, the present application further merges the adaptive time window feature vector library and the EWT feature tree before step S103 to form a multi-scale, multi-domain, multi-level device operation state feature library. The feature library contains not only static and dynamic features of the device on different time scales, but also oscillation and transient features of the device on different frequency scales.

Step S103, decomposing each data segment to obtain a plurality of component signals, calculating a complexity index corresponding to each component signal, and determining an abnormal complexity index from the plurality of complexity indexes.

Specifically, the application adopts component irregularity as a complexity index, and integrates the irregularity characteristics of the amplitude fluctuation, the frequency variation, the phase confusion and the like of the signal. The computation of component irregularities is based on the concept of distance matrix and recursive permutation entropy. And further, the complexity index with abnormality can be rapidly determined.

In some embodiments, the calculating the complexity index for each of the component signals includes: constructing a distance matrix corresponding to each component signal; performing recursive replacement on the distance matrix to obtain a plurality of replacement matrices corresponding to the distance matrix; calculating the permutation entropy corresponding to each permutation matrix; the complexity index of the component signal is obtained according to a plurality of the permutation entropies.

First, the present application constructs a distance matrix for each component signal. Each element of the distance matrix represents the distance between two points in time in the component signal. The distance may be measured using Euclidean distance, manhattan distance, or dynamic time warping distance. For example, for the component signal x (t), the element D _ij of its distance matrix D represents the distance between the time points t _i and t _j:

Next, the present application recursively permutes the distance matrix to generate a series of permutation matrices. The permutation matrix reflects the self-similarity and repeatability of the component signals over different time scales. In particular, the present application defines a permutation operator that reorders the rows and columns of the distance matrix simultaneously so that elements of the reordered distance matrix near the diagonal are as close as possible to the original distance matrix. The goal of the permutation operator is to minimize a difference metric, such as a sum of mean square error or absolute error, between the rearranged distance matrix and the original distance matrix. The replacement operator can be solved by an optimization method such as greedy search, dynamic programming or Hungary algorithm.

By recursive permutation, the present application can obtain a series of permutation matrices P ₁,P₂,……,P_K, where K represents the depth of the recursion. Each permutation matrix corresponds to a time scale reflecting the self-similarity of the component signals on that scale.

Then, the present application calculates permutation entropy for each permutation matrix. The permutation entropy measures the frequency and diversity of occurrence of different permutation modes in the permutation matrix. Specifically, the present application counts the number of occurrences of each permutation pattern in the permutation matrix and calculates their frequency distribution. The present application then uses shannon entropy to quantify the uncertainty and complexity of this frequency distribution. The calculation formula of the permutation entropy H _k is:

where n represents the order of the permutation matrix, p _i represents the frequency of occurrence of the ith permutation mode, n-! Representing the total number of all possible permutation patterns. The greater the permutation entropy, the higher the diversity of the permutation pattern in the permutation matrix, the higher the complexity of the component signal on the corresponding time scale.

Finally, the application carries out weighted average on the permutation entropy of all permutation matrixes to obtain the complexity index I of the component signal:

Where w _k represents the weight coefficient of the kth permutation matrix, reflecting the importance of the different time scales. The weight coefficient can be set empirically, or can be adaptively learned by a data-driven method.

The complexity index I quantifies the overall complexity of the component signal. The larger the I value is, the more remarkable the irregular characteristics such as amplitude fluctuation, frequency change, phase confusion and the like of the component signals are, and the more likely the equipment running state is abnormal or faulty.

For example, assume that the present application decomposes a vibration signal of a water chiller in an intelligent park to obtain a plurality of band component signals. For one of the component signals, the application calculates the distance matrix, the recursive permutation matrix and the permutation entropy, and finally obtains the complexity index i=0.85. The index value is higher, which indicates that the fluctuation and chaos of the component signal are stronger, and the abnormal abrasion or fatigue damage of a certain component (such as a compressor or a condenser fan motor) of the water chilling unit can be indicated.

By calculating the complexity of the component signals, the method can quantitatively evaluate the health level and the degradation degree of the running state of the equipment, and find potential abnormal or fault symptoms as soon as possible.

In some embodiments, said determining an abnormal complexity indicator among a plurality of said complexity indicators comprises: constructing a basic detector according to a plurality of preset abnormality detection algorithms; optimizing parameters of the base detector according to a grid search and cross-validation method; inputting each complexity index to the basic detector, and obtaining an anomaly score corresponding to the complexity index; and determining the complexity index with the anomaly score larger than a preset anomaly threshold as the anomaly complexity index.

Firstly, an integrated pool of anomaly detection algorithms is required to be constructed, and a plurality of classical anomaly detection algorithms are selected as basic detectors, such as a support vector machine (One-Class SVM), an isolated Forest (Isolation Forest), local anomaly factors (Local Outlier Factor) and the like. These algorithms can characterize the abnormal features of the data from different angles, such as density, distance, boundaries, etc. The application performs parameter tuning on each basic detector, and finds the optimal super-parameter combination on training data by using the techniques of grid search, cross verification and the like so as to achieve the optimal abnormality detection performance.

Next, the present application uses the trained base detector to score anomalies in the complexity index. For each complexity index sample, the application inputs it into all of the base detectors, resulting in a set of anomaly scores. The present application then employs an adaptive weight distribution strategy to dynamically adjust the weight coefficients of each detector based on its performance in validating the data.

Specifically, the present application uses a softmax function to translate the performance metrics (e.g., accuracy, recall, F1 score, etc.) of the detector into weights. The better performing detectors will get higher weights, while the worse performing detectors will be de-weighted. Through the self-adaptive weight distribution, the application can fully exert the advantages of each detector, inhibit the limitation of the detector and improve the robustness and generalization capability of anomaly detection.

The mathematical expression of the softmax function is as follows:

Where w _i represents the weight of the ith detector, p _i represents the performance index of the ith detector, such as the F1 score, T is a temperature parameter used to control the smoothness of the softmax function, p _j represents the performance index of the jth detector, and N is the total number of detectors.

After the weight distribution is completed, the application uses a weighted average mode to fuse the abnormal scores of the basic detector to obtain the final abnormal score of each complexity index sample. The higher the anomaly score, the more likely the sample is anomalous. The mathematical expression of the weighted average is as follows:

Where s represents the final anomaly score for the sample, w _i represents the weight of the ith detector, and s _i represents the anomaly score given by the ith detector.

In order to determine whether a sample is abnormal, an abnormality threshold is set in the present application. Unlike conventional fixed threshold methods, the present application employs an adaptive threshold adjustment strategy. Firstly, carrying out statistical modeling on abnormal scores in a normal state, and fitting a probability density function of the abnormal scores. Then, according to the expected anomaly ratio, the present application selects the right tail of the probability density function as the anomaly region, dynamically adjusts the threshold so that the area of the anomaly region is equal to the anomaly ratio.

Assuming that the anomaly score obeys normal distribution, the probability density function is:

where μ and σ are the mean and standard deviation of the anomaly scores, respectively. Given an anomaly ratio α, the anomaly threshold t should satisfy:

by numerically solving the above equation, the present application can obtain an adaptive anomaly threshold t.

Once the anomaly threshold is determined, the present application can make an anomaly determination for each of the complexity index samples. If the anomaly score of a sample exceeds an anomaly threshold, the present application marks it as an anomalous sample and identifies the corresponding component signal as an anomalous component signal. These abnormal component signals often carry critical information of the abnormal operation state of the equipment, and are important clues for fault diagnosis and predictive maintenance.

Step S104, fault information corresponding to the data segments is obtained according to the abnormal complexity index and the abnormal component signals corresponding to the abnormal complexity index, and a device fault ontology library is constructed according to the plurality of feature vectors and the fault information corresponding to each feature vector.

In the step, the abnormal fault information can be determined to construct a device fault ontology library by using the feature vector of the running state of the device extracted in the step and the identified abnormal complexity index and the abnormal component signal corresponding to the abnormal complexity index. The fault information in the ontology library comprises knowledge of equipment fault types, fault modes, fault symptoms and the like, and association relations and constraint conditions among the equipment. And further, diagnosis of real-time data can be completed quickly in the follow-up process.

First, the present application refers to the relevant standards of intelligent park equipment operation and maintenance, such as ASHRAE, IEC, etc., and consults experienced park engineers to define the core concept of the ontology library and the relationship between equipment by means of abnormal component signals and abnormal complexity indexes. These concepts include fault information for the equipment type (e.g., chiller, cooling towers, air conditioning ends, etc.), component names (e.g., compressor, condenser, evaporator, etc.), failure modes (e.g., refrigerant leakage, heat exchanger fouling, fan imbalance, etc.), failure symptoms (e.g., water supply temperature is too high, vibration is exacerbated, energy consumption is abnormal, etc.), severity (e.g., light, medium, heavy), etc. The application adopts the ontology description language OWL (Web Ontology Language) to formalize the concepts and the attributes of the fault information, and defines the semantic relation and the logic constraint of the fault information. For example, the present application defines a "chiller" as a type of equipment that has a "hasPart" (inclusive) relationship with "compressor," "condenser," "evaporator," etc.; while "compressor wear" is a failure mode that has a "hasSymptom" (apparent) relationship with the symptoms of "abnormal vibration", "increased noise", etc.

Next, the present application constructs an ontology concept describing different health states by using the feature vectors of the running states of the device obtained by the clustering algorithm in the above steps. For each cluster, the present application defines it as a health status concept such as "normal operation", "mild abnormality", "severe failure", etc. Each state concept contains statistical information such as a central feature vector, feature distribution parameters and the like of the cluster, and is used for describing typical performance of equipment features in the state. There is also a certain correspondence between these health status concepts and diagnostic concepts such as failure modes, failure symptoms, etc. For example, the present application may define a probability map matrix P, where element P _ij represents the conditional probability that symptom j was observed in state i. This matrix may be obtained by counting historical running data or may be given empirically by an expert. With the probability matrix, the application can express the association strength between different health states and fault symptoms in the ontology library, thereby providing probability basis for fault diagnosis based on symptoms. When constructing fault knowledge by using the abnormal component signals identified by the steps, the application focuses on the correspondence between the abnormal feature combination and the fault mode. By analyzing historical fault cases and expert diagnosis experience, the application summarizes typical abnormal characteristic modes of different fault modes and shows the characteristic modes in the form of ontology axiom. For example, for a typical failure "compressor wear" of a chiller, the present application may define the following axiom:

CompressorWear equivalentTo(Fault and(hasSymptom some AbnormalVibration)and(hasSymptom some NoiseRise)and(hasSymptom some PowerConsumptionIncrease)).

The axiom indicates that "compressor wear" is a failure that necessarily exhibits signs such as "abnormal vibration", "increased noise", "increased energy consumption", and the like. The fault definition axiom based on the abnormal characteristics can help the method to quickly match and diagnose the potential faults of the equipment, and reduce the problem investigation range. Of course, in practical applications, there may be some uncertainty in fault diagnosis based on abnormal features. The same fault may exhibit multiple combinations of anomalies, and a particular combination of anomalies may correspond to multiple possible causes of the fault. In order to improve the accuracy and reliability of diagnosis, the application introduces a probability representation in the semantic association of abnormal features with failure modes. By counting a large number of historical cases, the application can obtain a conditional probability matrix Q, wherein the element Q _ij represents the probability of judging as a fault mode j under the condition that an abnormal combination i is observed. For example, the present application may derive the following probability relationship:

P(CompressorWear|AbnormalVibration,NoiseRise,PowerConsumptionIncrease)＝0.85。

P(CompressorLeakage|AbnormalVibration,NoiseRise,PowerConsumptionIncrease)＝0.10。

P(CompressorLooseConnection|AbnormalVibration,NoiseRise,PowerConsumptionIncrease)＝0.05。

This means that when the present application observes three abnormal signs of "abnormal vibration", "noise increase", and "increase in energy consumption" of the compressor, it is judged that there is 85% probability that it is "compressor wear", 10% probability that it is "compressor leakage", and 5% probability that it is "compressor loosening". This probabilistic semantic association can help the present application quantify the uncertainty of fault diagnosis, reasonably evaluate the credibility of each possible cause, and thus make optimal decisions. In addition to building diagnostic knowledge of status, symptoms, faults, etc., the equipment fault ontology library also needs to represent knowledge of the association structure between campus equipment. These include topological connections of the devices in physical space, such as "cold water piping connects chiller units and air conditioning ends"; the equipment logically depends on functions, such as 'the refrigerating capacity of the tail end of the air conditioner depends on the water supply temperature of a water chilling unit'; and the propagation influence of equipment faults, such as high-pressure alarm of a water chilling unit caused by the fault of a cooling pump, and the like. The present application may use the object property mechanism of the ontology to characterize these structured association semantics, for example:

Chiller coolingWaterConnectedTo AirConditioningTerminal。

AirConditioning cooling Capacity dependsOn ChilledWater supplyTemperature。

Chiller hasAlarm HighPressure influencedBy CoolingPump hasFault。

These structured associative knowledge can help the present application analyze the health status of the device from local to global, infer the source and impact of the fault, optimize the scope and strategy of maintenance. For example, when the application diagnoses the fault of the water chilling unit, the refrigeration effect which possibly affects the tail end of the downstream air conditioner can be obtained through the association analysis, so that the application not only needs to overhaul the water chilling unit, but also needs to evaluate the influence on the indoor environment of the park, and if necessary, the application adopts the countermeasures of reducing the load in advance, scheduling the standby unit and the like.

Step S105, acquiring real-time measurement data of the equipment to be measured, extracting semantic features from the real-time measurement data, performing semantic matching and reasoning on the semantic features and the equipment fault ontology library, and acquiring a preliminary diagnosis result corresponding to the real-time measurement data.

In the step, the real-time measurement data with the measurement equipment are collected in real time, semantic features are extracted, and semantic matching and reasoning are carried out in the equipment fault ontology library constructed in the step S104. And identifying a fault mode of the equipment to be measured by combining the abnormal component signals identified in the steps and the corresponding diagnosis rules, and giving a preliminary diagnosis result.

In some embodiments, the extracting semantic features from the real-time measurement data, performing semantic matching and reasoning on the semantic features and the equipment fault ontology library includes: extracting the semantic features from the real-time measurement data according to a preset semantic feature mapping rule; matching the semantic features with the fault information in the equipment fault ontology library to obtain a symptom layer matching result; and reasoning according to the symptom layer matching result, acquiring a fault mode corresponding to the equipment to be measured, and completing semantic matching and reasoning of the semantic features and the equipment fault ontology library.

Specifically, the application firstly collects the operation parameters of park equipment such as temperature, pressure, flow, vibration and the like in real time through various sensors and monitoring systems. These parameters constitute a measurement data stream describing the real-time status of the device. In order to match these numerical measurement data with semantic knowledge in the ontology base, the present application requires feature extraction and semantic representation of the data. One common method is to convert the numerical features into semantic features, i.e. to describe the level, trend, etc. of the numerical values with a qualitative set of descriptors. For example, the present application may define the semantic feature mapping rules as follows:

the temperature T is more than 85 ℃ (hasTemperature, high); the temperature T is less than 60 ℃ and is hasTemperature, low. The pressure P is more than 1.2MPa (hasPressure, high); the pressure P is less than 0.8MPa (hasPressure, low); vibration amplitude V >2mm/s (hasVibration, abnormal); the vibration amplitude V is <0.5mm/s (hasVibration, normal).

High, low, abnormal, normal here are predefined semantic feature words that directly correspond to the fault symptom concepts in the ontology library. Through this mapping, the present application can convert real-time measurement data streams into semantic feature streams, such as (Chiller 1, hasTemperature, high) (Chiller 1, hasPressure, normal) (Chiller 1, hasVibration, abnormal).

Then, the application matches and infers the extracted semantic features with the equipment fault knowledge in the ontology library. This process can be divided into two layers, symptom-layer matching and fault-layer reasoning.

In symptom layer matching, the application firstly matches semantic features with fault symptom concepts defined in an ontology library. For example, the semantic features (Chiller, hasTemperature, high) described above may be matched to the ontology concept "HighTemperature", while (Chiller, hasVibration, abnormal) may be matched to the ontology concept "AbnormalVibration". By this matching, the present application can obtain a set of abnormal symptoms currently exhibited by the device, such as Chiller a 1 hasSymptom HighTemperature Chiller a hasSymptom AbnormalVibration a. These symptom layer matching results can be used to preliminarily determine the health status of the device. If several critical parameters of a device are abnormal, some kind of fault is likely to exist. However, symptoms alone are not sufficient to determine the type of fault, as the same symptom may correspond to multiple failure modes.

Therefore, the application also needs to make semantic reasoning at the fault layer. The purpose of the barrier layer reasoning is to infer the most likely failure mode of the device according to the matched abnormal symptoms by using the failure rules defined in the ontology library. Taking a water chiller as an example, the present application may define the following fault rules in the ontology library:

CompressorFailure equivalentTo(Fault and(hasSymptom some HighTemperature)and(hasSymptom some AbnormalVibration)and(hasSymptom some PressureIncrease)).

RefrigerantLeakage equivalentTo(Fault and(hasSymptom some PressureDecrease)and(hasSymptom some SubcoolingReduction)).

Here CompressorFailure denotes a compressor failure, REFRIGERANTLEAKAGE denotes a refrigerant leakage failure. They are all defined as a Fault and are fully conditioned by a set of necessary abnormal symptoms.

When the present application matches both of the HighTemperature and AbnormalVibration anomalies at the symptom level, the fault rule CompressorFailure can be triggered. Through semantic reasoning, the application can obtain the following diagnosis conclusion Chiller1 hasFault CompressorFailure. This indicates that chiller1, the last probability, has failed. However, if the application further matches PressureDecrease (pressure drop) and SubcoolingReduction (supercooling degree drop) symptoms at the symptom level, then REFRIGERANTLEAKAGE is triggered. At this point, there is conflict and uncertainty in fault diagnosis.

To resolve this uncertainty, the present application can distinguish and confirm the failure mode using the abnormal component signals identified in the previous steps. In the steps, the abnormal component which can reflect the fault characteristics is automatically extracted from the massive operation data of the equipment by the methods such as self-adaptive wavelet analysis and the like. These anomaly components have a more specific correspondence with a specific type of fault.

For example, the present application has found that the vibration signal at the time of a compressor failure may be significantly abnormally amplified in some frequency bands, while the pressure curve at the time of refrigerant leakage exhibits some typical falling pattern. These abnormal component signals can be used as strong evidence to confirm the type of fault.

Thus, after semantic reasoning yields possible fault candidates, the present application needs to further check whether the outlier component signals associated with these faults appear in the real-time data. If the abnormal frequency component associated with the compressor failure is detected and the abnormal pressure pattern associated with the refrigerant leakage does not occur, it is confirmed that the final failure diagnosis result is CompressorFailure instead of REFRIGERANTLEAKAGE.

The diagnosis method for fusing semantic reasoning and data anomaly analysis is established on the basis of two strong knowledge sources, namely the equipment fault ontology library and the anomaly characteristic library, and has higher accuracy and interpretability. On one hand, the semantic rules in the ontology library directly describe causal logic of faults and symptoms, and the diagnosis results have definite physical and mechanism interpretation; on the other hand, the abnormal component signal is used as a fault fingerprint of the data layer, so that ambiguity and uncertainty during matching of semantic rules can be effectively eliminated. The combination of the two enables the intelligent diagnosis system to integrate semantic knowledge (Semantic Knowledge) and data characteristics (DATA CHARACTERISTICS) like human expert, and obtain the most reliable and accurate diagnosis conclusion.

And S106, carrying out fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, and generating a measurement strategy according to the final diagnosis conclusion and the abnormal component signal to determine a measurement method of the equipment to be measured.

In the step, in the semantic matching and fault diagnosis, the application utilizes semantic rules in the equipment fault ontology library to preliminarily judge possible fault modes of equipment by combining equipment measurement data acquired in real time. For example, for a chiller, the present application may diagnose that it meets both the semantic conditions of the two failure modes, compressor failure (CompressorFailure) and refrigerant leakage (REFRIGERANTLEAKAGE). In this case, it is difficult to further judge the actual cause of the failure only by the semantic rule. And then adaptively generating a measurement strategy according to the confirmed final diagnosis conclusion and the identified abnormal component signal. And aiming at different fault types and abnormal components, dynamically adjusting the sampling frequency, the measuring point position and the measuring range of measurement, and realizing the dynamic allocation of measurement resources.

In order to improve the accuracy and reliability of diagnosis, the application needs to fuse and confirm the primary diagnosis result. The result fusion is to comprehensively utilize semantic matching results, abnormal component signals, equipment historical operation data and other multi-source sub-diagnosis results to verify and judge diagnosis assumptions so as to obtain a final fault diagnosis conclusion. This process can be seen as a post-processing and optimization of the primary diagnostic results.

In some embodiments, the sub-diagnostic result includes a plurality of semantic matching results, the outlier component signal corresponding to the semantic matching results, and the raw measurement data corresponding to the semantic matching results; and performing fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, wherein the fusion analysis comprises the following steps: calculating the matching degree of the abnormal component signal corresponding to the semantic matching result and the real-time measurement data; screening a plurality of semantic matching results according to the matching degree to obtain the similarity between the screened matching results and the original measurement data; and determining target similarity among the similarities, wherein the semantic matching result corresponding to the target similarity is the final diagnosis conclusion.

First, the present application makes full use of the abnormal component signals identified in the above steps to distinguish between different fault hypotheses. The application uses signal processing techniques such as self-adaptive wavelet analysis and the like to automatically extract the abnormal component signal which can reflect the fault characteristics from mass operation data of the equipment. These components tend to have a more specific correspondence with a particular fault type.

For example, compressor failure typically exhibits significant abnormal amplification at certain specific frequency bands of the vibration signal, while refrigerant leakage may cause some typical drop patterns in the pressure curve. These abnormal component signals can be used as strong evidence to identify the type of fault.

Therefore, after obtaining the preliminary semantic diagnosis, the present application further checks whether the abnormal component signals associated with these faults hypotheses appear in the real-time data. The application may define an abnormal component matching metric function M (F _i, D) representing the degree of matching of the abnormal component signal of fault F _i in the current data D. The function may be constructed based on correlation of signals, distance metrics, statistical tests, or the like. For example, the present application may characterize the degree of matching by the energy duty cycle of the anomaly component:

Where S _i represents the abnormal component signal set of fault F _i, E _k represents the energy of the kth abnormal component, N is the total number of data components. This index characterizes the proportion of the energy of the fault-related anomaly component in the overall data, and the higher the matching degree, the more likely the current data contains the characteristics of the fault.

By utilizing the abnormal component matching information, the application can carry out preliminary screening on the result of the volunteer medical consultation break. For example, if the abnormal component match for a compressor failure is significantly higher than a refrigerant leak, the present application may infer that the compressor failure is a more likely cause. However, if the anomaly component of both faults does not manifest itself significantly, then other fault possibilities or data re-acquisition may need to be considered.

In addition to the abnormal component information, the application further confirms the diagnosis result by combining the historical operation data of the equipment. The historical data of the equipment contains abundant case knowledge of fault occurrence, development and repair, and has important reference value for judging the nature and severity of the current fault.

For example, if a compressor failure has occurred multiple times in the past year for a particular chiller, and the symptoms before the failure are consistent with the current performance, then the confidence of the compressor failure diagnosis may be increased. Conversely, if the unit has a history of frequent refrigerant leaks and the compressor is operating steadily at all times, the likelihood of refrigerant leaks may be greater.

The application can use Case-based reasoning (Case-based Reasoning) technology to measure the similarity of the current fault symptoms and the historical fault cases, find out the most similar cases and take the diagnosis result as the reference of the current faults. The case similarity can be modeled by a weighted distance function or a learning ordering model and other methods by considering various factors such as symptom characteristics, measurement parameters, environmental conditions, equipment model and the like.

In addition, the historical operation data can be used for constructing a degradation trend model of equipment and evaluating the evolution stage and risk level of the current fault. For example, by tracking the long-term trend of the key parameters of the equipment, the application can judge whether the current fault is in an early stage or is close to the critical point of failure, thereby determining the urgency of the diagnosis result and the priority of the countermeasure.

The application can fuse and confirm the primary diagnosis result by comprehensively considering the results of semantic rule matching, abnormal component detection and historical data analysis. One common fusion strategy is a weighted voting method, i.e. different weights are given to the diagnosis results according to the credibility of different information sources, and then weighted summation is carried out to obtain the final comprehensive diagnosis conclusion.

Assuming that the fault probability distributions obtained by the semantic diagnosis, the abnormal component diagnosis and the historical data diagnosis are P _s(F_i),P_a(F_i),P_h(F_i respectively and the weights of the fault probability distributions are w _s,w_a,w_h respectively, the final fusion diagnosis probability P (F _i) can be expressed as:

P(F_i)＝w_sP_s(F_i)+w_aP_a(F_i)+w_hP_h(F_i).

The application can select the fault mode with the highest probability as the final diagnosis conclusion, and takes the probability value as the confidence level of diagnosis:

Where F ^* is the failure mode of the final diagnosis and C is the diagnostic confidence.

In some embodiments, the measurement strategy includes a sampling frequency, a site location, and a measurement range; said generating a measurement strategy based on said final diagnostic conclusion and said abnormal component signal, comprising: determining the position and the measuring range of the measuring point according to the final diagnosis conclusion; and determining the sampling frequency according to the frequency range of the abnormal component signal.

Specifically, the core idea of adaptive measurement strategy generation is to adopt different measurement methods for different types of faults and abnormal characteristics, and obtain the most valuable fault information with the minimum measurement cost. The measuring method mainly comprises three aspects of sampling frequency, measuring point position and measuring range.

The sampling frequency refers to the number of times measurement data is acquired per unit time. In general, the higher the sampling frequency, the finer and complete the data obtained, but at the same time also means a greater data transmission and storage overhead. Therefore, selecting an appropriate sampling frequency requires a tradeoff between data quality and resource consumption. The present application knows that the performance characteristics of different fault types in the time domain and the frequency domain are different. For example, certain faults may develop slowly, exhibiting long-term trend changes, such as equipment wear, aging, etc.; while other faults may occur rapidly, manifested as transient pulse signals such as component breaks, shorts, etc. For trend slowly varying faults, the application can adopt relatively lower sampling frequency, such as sampling once per minute or hour, so that the overall characteristics of the faults can be captured, and the data volume can be reduced. Whereas for transient type sharp faults, the present application requires the use of a higher sampling frequency, such as multiple samples per second or millisecond, to avoid missing critical fault moments.

In addition, the application may further optimize the sampling frequency based on the identified abnormal component signal. By returning to the steps, the application automatically extracts the abnormal component which can reflect the fault characteristics from the massive operation data of the equipment by using signal processing technologies such as wavelet analysis and the like. These components correspond to the energy distribution of the fault over different frequency bands. For example, the application discovers that the fault of a fan motor is mainly reflected in two frequency bands of 125Hz and 300 Hz. According to the nyquist sampling theorem, the sampling frequency of the present application needs to be greater than two times, i.e., greater than 600Hz, in order to be able to reconstruct the signals of both frequency bands accurately. Therefore, the application can adaptively determine the optimal sampling frequency according to the frequency distribution of the abnormal component signals, and neither key fault information is missed nor resource waste is caused by over-sampling.

The measuring point position refers to the installation position of the sensor and the instrument. Different measuring point arrangement schemes have great influence on the coverage range and the sensitivity of fault monitoring. Conventional site placement is often empirically or normative and is difficult to accommodate for changes in equipment architecture and failure modes. The self-adaptive measurement strategy generation method dynamically optimizes the selection and layout of the measuring points according to the specific fault type and the specific part which are diagnosed. For example, if the diagnosis indicates that a pump may have a bearing failure, the present application requires vibration and temperature sensors to be added near the bearing, focusing on monitoring the abnormal state of the bearing. If the pump blade is suspected to have a problem, pressure and flow sensors should be arranged at the inlet and the outlet of the impeller, and the working condition change of the blade is closely focused.

Meanwhile, by means of the extracted abnormal component signals, the application can further optimize the position of the measuring point. The abnormal component signal reflects not only the frequency characteristics of the fault but also the spatial distribution information thereof. For example, by modal analysis of vibration signals, the present application has found that abnormal vibration of a certain compressor is mainly concentrated at the front end of the second cylinder. This suggests that there may be a fault hazard in this location, requiring major monitoring. Therefore, the application can automatically identify the high risk areas of faults according to the spatial distribution diagram of the abnormal component signals, and correspondingly increase the measuring point density of the areas so as to improve the monitoring precision and reliability.

The measurement range refers to the range of the sensor and meter. The reasonable measurement range is required to fully cover the normal working range of equipment, and a sufficient margin is reserved for coping with possible abnormal fluctuation. However, if the measurement range is set too large, the sensitivity and resolution of the measurement are reduced; if the measurement range is too small, the extreme value caused by the fault may not be captured. Therefore, the self-adaptive measurement strategy generation method dynamically adjusts the measurement range according to the diagnosed fault severity and the abnormal amplitude, thereby avoiding the measurement blind area and improving the measurement precision.

For example, if the diagnostic result indicates that a device has only a slight initial fault and its abnormal parameter value is still in the vicinity of the normal range, the present application may properly narrow the measurement range, focusing more resolution around the normal value, so as to find minor changes in the fault early. Conversely, if the diagnostic discovers that the device is in a severely malfunctioning state, many of its parameter values are significantly outside of the normal range, the measurement range needs to be correspondingly enlarged to avoid saturation or truncation of the measurement. The adjustment amplitude here may refer to the numerical distribution of the abnormal component signal extracted in the above step. For example, the present application can form an adaptive measurement range with 1.5 times the maximum value and 0.5 times the minimum value as the new upper and lower measurement limits for the abnormal vibration amplitude component.

The following description of the process of adaptive measurement strategy generation is presented by way of a specific example.

Assuming that in the steps, the application confirms that a certain centrifugal chiller has early failure of the compressor bearings, and the confidence coefficient is 80%. Meanwhile, in the above steps, the abnormal component signal of the present application, which identifies the fault, is mainly concentrated in the high-frequency vibration signal of 1kHz-5kHz, and the abnormal peak value of the vibration amplitude is 5g.

According to the above diagnosis results, the present application generates the following adaptive measurement method:

1. sampling frequency-according to the frequency range of abnormal component signal, the sampling frequency of vibration sensor of compressor bearing position is raised to 10kHz so as to ensure that the fault signal of 1kHz-5kHz can be completely captured. For other measuring points, the original sampling frequency of 1kHz is maintained unchanged, so that data resources are saved.

2. And the measuring point positions are respectively provided with a vibration sensor at the driving end and the non-driving end of the compressor bearing to form a three-point one-line measuring point layout so as to comprehensively monitor the vibration state of the bearing. Meanwhile, a temperature sensor is additionally arranged near the bearing, and abnormal heating condition of the bearing is monitored. The other station positions remain unchanged.

3. Measuring range is that the measuring range of the vibration sensor of the compressor bearing is adjusted to be +/-7.5 g from the original +/-10 g, so that the resolution of early tiny abnormal vibration of the bearing is improved. Correspondingly, the measuring range of the bearing temperature sensor is adjusted to 30-70 ℃ from 0-100 ℃ so as to monitor the temperature rise process of the bearing more finely. The measurement range of the other sensors remains unchanged.

Through the adjustment of the self-adaptive measurement strategy, the method can capture and monitor early fault symptoms of the compressor bearing more specifically and more efficiently, and provide high-quality data support for subsequent trend prediction and maintenance decision.

In some embodiments, after said generating a measurement strategy to determine a measurement method for said device to be measured based on said final diagnostic conclusion and said abnormal component signal, further comprising: determining a fault early warning threshold corresponding to the equipment to be measured according to the measurement strategy and the final diagnosis conclusion; and when the measurement parameter corresponding to the real-time measurement data is larger than the fault early warning threshold value, generating an early warning signal to finish early warning and real-time monitoring of the equipment to be measured.

The fault early warning threshold is a parameter limit for triggering a fault early warning signal when the measured parameter reaches or exceeds a certain critical value. Reasonable early warning threshold values need to be balanced between the fault detection rate and the false alarm rate, so that fault symptoms are found as early as possible, and excessive invalid alarms are avoided. Conventional fixed threshold methods often rely on experience or criteria to set a static threshold, such as vibration speeds in excess of 10mm/s, temperatures above 80 c, etc. However, the method of 'one-cut' ignores the influence of factors such as the running working condition, the health state, the environmental condition and the like of equipment, so that the early warning effect is poor. For example, a vibration speed threshold of 10mm/s may be too loose for low speed devices and too tight for high speed devices. Therefore, the application needs to adaptively determine and adjust the early warning threshold according to the actual operation characteristics and failure modes of the equipment.

Specifically, the self-adaptive fault early warning threshold value determining method mainly comprises three steps of threshold value initialization based on historical data, threshold value self-adaptive optimization based on online data and threshold value constraint adjustment based on expert experience.

Firstly, the application utilizes historical operation and fault data to perform initialization setting on the early warning threshold value. The historical data mainly comprises two parts, namely normal operation data and fault case data. The application carries out statistical analysis on the historical normal value of each measured parameter to obtain the mean value mu and the standard deviation sigma. According to the 3 sigma principle, the application can take mu + -3 sigma as the initial normal range of the parameter, and values outside this range are regarded as abnormal or faulty. For example, if the historical data of the temperature sensor is 60 ℃ in average and 5 ℃ in standard deviation, the initial normal range is [45 ℃,75 ℃), and the temperature value higher than 75 ℃ or lower than 45 ℃ triggers early warning.

However, the threshold values obtained from normal data alone tend to be relatively conservative, and may miss some early or implicit signs of failure. Therefore, the application also needs to analyze the historical fault case data to find out the abnormal change mode and the critical value of the measured parameter under each fault mode. For example, through statistical analysis, the present application found that bearing wear failure of a certain type of motor generally starts to manifest itself when its vibration speed reaches around 6mm/s, and is rapidly deteriorated when the vibration speed exceeds 8 mm/s. Therefore, the application can take 6mm/s as the early warning threshold value of the motor bearing fault and 8mm/s as the serious warning threshold value.

The threshold values obtained by analyzing the normal data and the fault data are compared and integrated, and the application can obtain a group of initial fault early warning threshold values. These thresholds cover the main measured parameters and failure modes, reflecting the operational boundaries of the device under different health conditions.

For example, due to limitations in historical data and constant changes in device status, the initial threshold may not be well adapted to the current operating conditions. Therefore, the application also requires adaptive optimization of the threshold value on-line in real time. This uses the adaptive measurement strategy generated in the above steps.

According to the real-time fault diagnosis result and the abnormal characteristics of the equipment, the application dynamically adjusts the sampling frequency, the measuring point position and the measuring range of measurement, and generates a set of self-adaptive measuring parameters. These parameters reflect the health status and degradation trends of the device in real time. The application can take the adaptive parameters as the basis and input of threshold optimization.

For example, if the diagnosis result indicates that a slight initial fault occurs in the equipment, the monitored parameter starts to show an abnormal trend, but the initial early warning threshold is not reached, the early warning threshold of the corresponding parameter can be properly reduced, so that early fault early warning is realized. When the monitored parameter is continuously deteriorated and exceeds the initial early warning threshold, the early warning level can be further improved until the alarm threshold is triggered.

For another example, if the measurement strategy increases the vibration sampling frequency of the bearing from 1kHz to 10kHz, it is stated that the dither characteristics of the bearing may contain important fault information. At this time, the application not only focuses on the change of the vibration speed, but also focuses on analyzing the indexes such as the vibration acceleration and the peak factor of the high frequency band, and correspondingly sets and optimizes the early warning threshold value.

In order to quantitatively describe the relationship between the measurement parameters and the early warning threshold, the application can establish an adaptive threshold optimization model. The model takes a current measurement parameter value x _t and a historical early warning threshold y _t-1 as inputs, and outputs an optimized new early warning threshold y _t. The general form of which can be expressed as y _t＝f(x_t,y_t-1, θ).

Wherein f is a threshold optimization function, θ is a super parameter of the optimization function, and the super parameter can be obtained by training from historical data through methods such as machine learning. Common threshold optimization functions include weighted moving average, exponential smoothing, adaptive control limiting, and the like. For example, the exponential smoothing method may weight average the current parameter value with a historical threshold, the weight being determined by a smoothing factor α:

y_t＝αx_t+(1-α)y_t-1。

By adjusting the smoothing factor, the application can control the sensitivity of the threshold to current parameter variations. The larger the alpha is, the faster the threshold value changes with the current parameter; the smaller α, the more stable the threshold. The value of the smoothing factor can be adaptively adjusted according to the development speed of faults, the measured noise level and other factors.

In addition to the data-driven threshold optimization method, the application also needs to fully utilize expert experience to constrain and fine tune the threshold. The deep engineer and maintainer accumulate a great deal of equipment fault diagnosis and preventive maintenance experience, and master the abnormal modes and critical values of a plurality of key measurement parameters. The application can convert the experiences into threshold constraint conditions through knowledge acquisition, rule extraction and other methods, and is used for guiding and checking the optimization result of the model.

For example, expert experience may indicate that for a certain type of pump, cavitation is highly likely to occur once the peak-to-peak pressure pulsation exceeds 20% of rated pressure, resulting in impeller and shaft seal damage. Therefore, the application can set a constraint rule that the pressure pulsation early warning threshold value of the pump is not higher than 20% of Yu Eding pressure. If the threshold value obtained by optimizing the model violates the rule, the application needs to carry out manual intervention and correction.

In addition, the expert can also subjectively adjust the early warning threshold values of different measurement parameters according to factors such as safety margin, fault consequences, overhaul difficulty and the like of the equipment. For example, for critical components that may cause serious accidents or be difficult to repair once they fail, an expert may recommend more stringent and conservative warning thresholds to discover hidden hazards as soon as possible, improving safety. For those general components with slight fault consequences or easy replacement, the expert may properly relax the early warning threshold to reduce the false alarm probability and save the maintenance cost.

By integrating the steps, the application can finally obtain a set of dynamically optimized fault early warning threshold system. The threshold system is based on historical data analysis, on-line measurement parameters are used as optimization basis, expert experience is used as constraint condition, and the change of the health state of the equipment can be tracked in a self-adaptive manner, so that early, accurate and comprehensive fault early warning is realized.

The application will now be described with a specific example of the effect of this method.

A centrifugal compressor in a chemical industry park is generated by the provided self-adaptive measurement strategy, and the fault of the centrifugal compressor is identified to be mainly related to the vibration leakage of the impeller and the bearing. When the initial early warning threshold value is set, the normal range of the impeller vibration speed is 0-5mm/s, and the normal range of the bearing vibration speed is 0-3mm/s according to the historical data analysis. Meanwhile, the fault case data show that when the vibration speed of the impeller exceeds 10mm/s, blade breakage often occurs; when the bearing vibration speed exceeds 6mm/s, severe wear has occurred in most cases. Thus, the application takes 10mm/s and 6mm/s as the initial alarm threshold for the impeller and bearing vibration speeds, respectively.

In actual operation monitoring, the self-adaptive measurement module finds that the vibration speeds of the impeller and the bearing of the compressor have rising trend, and the vibration speeds are respectively increased from 3mm/s and 2mm/s to 7mm/s and 4mm/s. Although the initial alarm threshold has not been reached, the acceleration rate has deviated significantly from the normal range. Accordingly, the threshold optimization model adjusts the early warning threshold of the impeller vibration speed from 10mm/s to 9mm/s and adjusts the early warning threshold of the bearing vibration speed from 6mm/s to 5mm/s. Meanwhile, the model also constrains the threshold according to the empirical rule, the threshold of the impeller vibration speed is required to be not lower than 8mm/s, and the threshold of the bearing vibration speed is required to be not lower than 4mm/s so as to avoid excessive false alarm.

In the next week, the compressor wheel vibration speed further deteriorated to 8.5mm/s, triggering the early warning signal. The operator on duty in time stops the compressor and inspects, discovers that a blade has cracks and has the risk of falling off. By replacing the blades and correcting the dynamic balance, the vibration speed of the compressor is restored to the normal level, and serious accidents such as blade breakage and the like are avoided. Meanwhile, although the vibration speed of the bearing is increased, the vibration speed is kept below 5mm/s, and no early warning is triggered. Inspection shows that the bearing is well lubricated and slightly worn, and the bearing is not needed to be replaced temporarily. The threshold optimization model can better weigh the fault detection rate and the false alarm rate, so that key faults can be found in time, excessive alarm is avoided, and maintenance resources are reasonably distributed.

Illustratively, after the generating the early warning signal and completing the early warning and the real-time monitoring of the device to be measured, the method further includes: determining an equipment maintenance strategy of the equipment to be measured according to the fault early warning threshold, the measurement strategy and the final diagnosis conclusion; the maintenance strategy comprises a maintenance period, a maintenance mode and maintenance resource allocation; acquiring health information, fault risk information and early warning grade information of the equipment to be measured; and optimizing the maintenance period, the maintenance mode and the maintenance resource allocation according to the health information, the fault risk information and the early warning level information.

Specifically, the core of the device maintenance policy optimization is to build a multi-objective optimization model. The model takes the health state, fault risk and maintenance cost of the equipment as main decision variables, and the reliability, availability and life cycle cost of the equipment as key optimization targets. Meanwhile, the model takes a fault diagnosis result, measurement strategy parameters and an early warning threshold value as important constraint conditions, so that the optimized maintenance strategy is ensured to meet the performance requirements of equipment monitoring diagnosis. By solving the multi-objective optimization problem, the application can obtain the optimal time plan and the optimal maintenance mode combination for equipment maintenance and the most reasonable maintenance resource allocation, thereby maximizing the life cycle value of the equipment.

The optimization model aims at maximizing the reliability R (s, λ, C) and availability a (s, λ, C) of the device under the health state s, failure rate λ and maintenance cost C, while minimizing its lifecycle cost C (s, λ, C). Constraints of the model include that the health status s of the device must belong to a predefined set of health statuses H ₁,H₂,...,H_n; the failure rate lambda of the device must not exceed the failure rate threshold lambda _T; the maintenance cost c of the equipment must be within an acceptable cost interval c _L,c_U; the accuracy f _D(s) of fault diagnosis is larger than the minimum accuracy requirement alpha _D; the fault pre-warning accuracy f _P (s, lambda) is less than the maximum false positive rate tolerance beta _P.

In order to solve the multi-objective optimization model, the application firstly needs to accurately estimate the current health state of the equipment according to the fault diagnosis result of the steps. The accuracy of the fault diagnosis directly affects the reliability of the health status estimation. Then, the application obtains the real-time monitoring data of the equipment in different health states by utilizing the self-adaptive measurement strategy of the steps. By carrying out statistical modeling and machine learning analysis on the monitoring data, the application can establish a quantitative relation model between the health state and the equipment failure rate and between the health state and the key measurement parameters. For example, the present application may use a Cox proportional hazards model to describe the dependency between the health of a device and its failure rate function:

λ(t|s)＝λ₀(t)exp(β₁x₁+β₂x₂+…+β_mx_m)。

Where λ (t|s) represents the failure rate function for a given state of health s, λ ₀ (t) is the base failure rate function, x ₁,x₂,…,x_m is the covariate affecting the failure rate, such as measured parameters of temperature, pressure, vibration, etc., and β ₁,β₂,…,β_m is the corresponding regression coefficient. By carrying out parameter estimation on the historical operation data and the fault data, the application can obtain the specific form of the fault rate function.

The present application is to make full use of the fault pre-warning threshold values in the above steps to evaluate the current health status and risk of faults of the device. If the monitored parameters exceed the pre-warning threshold, this means that the device has significant signs of abnormality, and there is a high likelihood that a functional failure will occur in a short period of time, requiring maintenance or replacement to be scheduled as soon as possible. On the contrary, if the monitoring parameter is far lower than the early warning threshold value, the equipment running state is good, the maintenance period can be properly prolonged, and unnecessary shutdown and overhaul are reduced. The application can define a risk assessment function R (s, lambda) comprehensively considering the health state and the fault rate, and quantitatively judge the current risk level of the equipment:

R(s，λ)＝f(s)·g(λ)。

Wherein f(s) and g (λ) are monotonically increasing functions with respect to health and failure rate, respectively. The larger the value of the risk assessment function, the worse the health status of the device, the higher the failure rate and the stronger the urgency of maintenance.

After the results of health state estimation, fault rate prediction and risk assessment are obtained, the maintenance strategy optimization model can be built. The application converts the maintenance decision problem into a multi-objective optimization problem. The optimized decision variables mainly comprise two major categories, namely, maintenance time schedule, namely, when preventive maintenance and equipment update are carried out; and secondly, the maintenance mode and intensity, namely what type of maintenance activity is adopted and how much maintenance resource is input. The present application can combine these decision variables into one maintenance policy vector m:

m＝(t₁,t₂,…,t_k,d₁,d₂,…,d_k,r₁,r₂,…,r_k).

Where t _i denotes the time of the ith maintenance, d _i denotes the mode of the ith maintenance, r _i denotes the resource allocation of the ith maintenance, i e [1, k ], and k is the total number of maintenance times.

The object of the present application is to find an optimal combination of maintenance time schedule and maintenance pattern throughout the life cycle of the device such that the reliability, availability of the device is maximized and the total maintenance costs are minimized. The present application can build corresponding objective functions to quantify these optimization objectives. For example, the reliability function R (t|m) may be found based on the failure rate function λ (t|s) and the maintenance policy m:

Where s _m (x) represents the health of the device at time x under maintenance policy m. Similarly, the present application can build a calculation model of the availability function a (t|m) and the cost function C (t|m).

Finally, the application designs a solving method based on an improved Particle Swarm Optimization (PSO) algorithm by utilizing a multi-objective Optimization intelligent algorithm, and solves the maintenance strategy Optimization model to obtain an optimal equipment maintenance plan and resource allocation scheme.

First, the present application converts the maintenance decision problem into a discrete combination optimization problem. The application distributes a binary decision variable for each possible maintenance time point of each equipment, wherein a variable value of 0 indicates that maintenance is not performed, and a value of 1 indicates that maintenance is performed. In this way, the maintenance decision variables of all devices together form a high-dimensional discrete solution space, each solution vector representing a possible maintenance time plan.

The present application then applies a PSO algorithm to the search optimization of the discrete solution space. Based on the classical PSO algorithm, the application carries out the following key improvements:

1. Particle encoding-each particle is encoded as a binary matrix, the rows of the matrix representing the device and the columns representing the time window. The matrix element takes a value of 0 or 1, which respectively indicates whether the corresponding device is maintained in the time window.

2. Particle initialization, namely, the initial population comprises randomly generated particles and heuristic particles generated according to the field knowledge such as equipment importance, fault risk, maintenance period and the like are mixed, so that the quality and diversity of an initial solution are improved.

3. And (3) evaluating the particles, namely predicting multiple target values such as equipment reliability, maintenance cost and the like under the maintenance plan represented by each particle through simulation tools such as an equipment degradation model, a maintenance effect model and the like, and carrying out normalization weighting summation to obtain the comprehensive fitness value of the particles.

4. Particle updating, namely introducing self-adaptive inertia weight and learning factors to enable an algorithm to present different global exploration and local development capacities at different stages. And meanwhile, reasonably repairing particles which violate resource constraint and task coordination constraint through an intelligent constraint processing mechanism.

5. And (3) field searching, namely periodically importing a high-quality maintenance mode generated according to equipment fault diagnosis results and maintenance experience into the population, and performing cross operation with common particles to accelerate the evolution and convergence of the population.

By iteratively executing the steps, the particle swarm is continuously updated, and finally the pareto optimal solution set of the problem is converged. The application selects the solution with the highest comprehensive fitness value from the solution set as the final equipment maintenance strategy, and generates an operable maintenance operation instruction book and a spare part purchasing plan according to the maintenance plan and the resource allocation scheme corresponding to the strategy.

The application takes a centrifugal water chilling unit of an intelligent park as an example to describe the application flow of equipment maintenance strategy optimization.

Assuming a centrifugal chiller set in the campus, it is responsible for providing chilled water to the air conditioning system and process equipment of the campus. According to the application, an advanced online monitoring system is arranged on each water chilling unit, and key parameters such as vibration, temperature, pressure, current and the like of the units are collected and analyzed in real time. Through the fault diagnosis model of the steps, the application discovers that the motor bearing of the No. 1 water chilling unit is in a slight abrasion state, the health index is 0.8, and the estimated fault rate is 0.2 times/year. The self-adaptive measurement strategy in the steps shows that the vibration level of the water chilling unit bearing is increased, but the early warning threshold value is not exceeded. Through risk assessment, the application judges that the current comprehensive risk level of the No. 1 water chilling unit is 0.35, and belongs to low and medium risk levels.

Aiming at the diagnosis result and risk assessment of the No. 1 water chilling unit, the application preliminarily establishes a maintenance strategy that routine maintenance is carried out on the unit every 3 months in the future 1 year, including engine oil replacement, filter cleaning, safety protection device verification and the like; performing special detection every half year, and focusing on the vibration state of the bearing; and the overhaul or replacement of the bearing is not scheduled temporarily. Then, the maintenance strategy is input into a multi-objective optimization model, the minimum cost and the highest reliability of the life cycle of the unit are taken as optimization targets, and the fault diagnosis accuracy rate is more than 95% and the like are taken as constraint conditions, and the genetic algorithm is utilized for solving.

The optimal annual maintenance strategy of the No. 1 water chilling unit is obtained through multiple rounds of iterative optimization, wherein the unit is routinely maintained in the 2 nd, 5 th, 8 th and 11 th months in the next 1 year, and 16 working hours are input each time; performing special detection once in the 6 th month, and inputting 24 working hours; the bearing was subjected to a small repair at 10 months, the grease was replaced, and 32 man-hours were put in. Under the optimal strategy, the annual reliability of the No. 1 water chilling unit can reach more than 98%, and the total maintenance cost in the life cycle of 20 years can be reduced by about 15%. Compared with the original fixed-period maintenance strategy, the self-adaptive maintenance strategy ensures the high-reliability operation of the unit and simultaneously reduces the maintenance cost and the resource consumption to the greatest extent.

In order to execute the device measurement method facing the intelligent park corresponding to the method embodiment, corresponding functions and technical effects are realized. Referring to fig. 2, fig. 2 shows a block diagram of an apparatus measurement device 200 for an intelligent park according to an embodiment of the present application. For convenience of explanation, only the portions related to this embodiment are shown, and the device measurement apparatus 200 for an intelligent park according to the embodiment of the present application includes:

the data acquisition module 201 is configured to acquire a plurality of original measurement data of a preset device in the smart park, perform data cleaning and format conversion on the original measurement data, and acquire standardized target data.

The time sequence segmentation module 202 is configured to perform time sequence segmentation on the target data, obtain a plurality of data segments, and extract a feature vector of each data segment.

The complex computing module 203 is configured to decompose each data segment, obtain a plurality of component signals, compute a complexity index corresponding to each component signal, and determine an abnormal complexity index from the plurality of complexity indexes.

The fault construction module 204 is configured to obtain fault information corresponding to the data segment according to the abnormal complexity index and an abnormal component signal corresponding to the abnormal complexity index, and construct an equipment fault ontology library according to a plurality of feature vectors and fault information corresponding to each feature vector.

The preliminary diagnosis module 205 is configured to obtain real-time measurement data of a device to be measured, extract semantic features from the real-time measurement data, perform semantic matching and reasoning on the semantic features and the device fault ontology library, and obtain a preliminary diagnosis result corresponding to the real-time measurement data.

And a final diagnosis module 206, configured to perform fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results, obtain a final diagnosis conclusion, and generate a measurement policy according to the final diagnosis conclusion and the abnormal component signal, so as to determine a measurement method for the device to be measured.

The smart-park oriented device measurement apparatus 200 may implement the smart-park oriented device measurement method of the method embodiment. The options in the method embodiments described above are also applicable to this embodiment and will not be described in detail here. The rest of the embodiments of the present application may refer to the content of the above method embodiments, and in this embodiment, no further description is given.

Fig. 3 is a schematic structural diagram of a computer device according to an embodiment of the present application. As shown in fig. 3, the computer device 3 of this embodiment includes: at least one processor 30 (only one is shown in fig. 3), a memory 31 and a computer program 32 stored in the memory 31 and executable on the at least one processor 30, the processor 30 implementing the steps in any of the method embodiments described above when executing the computer program 32.

The computer device 3 may be a smart phone, a tablet computer, a desktop computer, a cloud server, or other computing devices. The computer device may include, but is not limited to, a processor 30, a memory 31. It will be appreciated by those skilled in the art that fig. 3 is merely an example of the computer device 3 and is not meant to be limiting as the computer device 3, and may include more or fewer components than shown, or may combine certain components, or different components, such as may also include input-output devices, network access devices, etc.

The Processor 30 may be a central processing unit (Central Processing Unit, CPU), the Processor 30 may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL processors, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), off-the-shelf Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

The memory 31 may in some embodiments be an internal storage unit of the computer device 3, such as a hard disk or a memory of the computer device 3. The memory 31 may in other embodiments also be an external storage device of the computer device 3, such as a plug-in hard disk, a smart memory card (SMART MEDIA CARD, SMC), a Secure Digital (SD) card, a flash memory card (FLASH CARD) or the like, which are provided on the computer device 3. Further, the memory 31 may also include both an internal storage unit and an external storage device of the computer device 3. The memory 31 is used for storing an operating system, application programs, boot loader (BootLoader), data, other programs etc., such as program codes of the computer program etc. The memory 31 may also be used for temporarily storing data that has been output or is to be output.

In addition, the embodiment of the present application further provides a computer readable storage medium, where a computer program is stored, where the computer program is executed by a processor to implement the steps in any of the above-mentioned method embodiments.

Embodiments of the present application provide a computer program product which, when run on a computer device, causes the computer device to perform the steps of the method embodiments described above.

In several embodiments provided by the present application, it will be understood that each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved.

The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application may be embodied essentially or in a part contributing to the prior art or in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device to perform all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The foregoing embodiments have been provided for the purpose of illustrating the general principles of the present application, and are not to be construed as limiting the scope of the application. It should be noted that any modifications, equivalent substitutions, improvements, etc. made by those skilled in the art without departing from the spirit and principles of the present application are intended to be included in the scope of the present application.

Claims

1. A method for measuring equipment oriented to an intelligent park, the method comprising: acquiring a plurality of original measurement data of preset equipment in an intelligent park, and performing data cleaning and format conversion on the original measurement data to acquire standardized target data;

Performing time sequence segmentation on the target data to obtain a plurality of data fragments, and extracting feature vectors of each data fragment;

Decomposing each data segment to obtain a plurality of component signals, calculating a complexity index corresponding to each component signal, and determining an abnormal complexity index in the plurality of complexity indexes; the determining an abnormal complexity index from a plurality of complexity indexes comprises: constructing a basic detector according to a plurality of preset abnormality detection algorithms; optimizing parameters of the base detector according to a grid search and cross-validation method; inputting each complexity index to the basic detector, and obtaining an anomaly score corresponding to the complexity index; determining the complexity index with the anomaly score larger than a preset anomaly threshold as the anomaly complexity index; wherein the probability density function of the anomaly score is: The probability density function is f(s), mu and sigma are the mean value and standard deviation of anomaly scores, alpha is a given anomaly proportion, and the anomaly threshold t should satisfy: Dynamically adjusting an abnormal threshold by selecting the tail part on the right side of the probability density function as an abnormal region, so that the area of the abnormal region is equal to the abnormal proportion, and realizing an adaptive threshold adjustment strategy for the abnormal threshold;

acquiring real-time measurement data of equipment to be measured, extracting semantic features from the real-time measurement data, carrying out semantic matching and reasoning on the semantic features and the equipment fault ontology library, and acquiring a preliminary diagnosis result corresponding to the real-time measurement data;

And carrying out fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, and generating a measurement strategy according to the final diagnosis conclusion and the abnormal component signal so as to determine a measurement method of the equipment to be measured.

2. The method of claim 1, wherein said decomposing each of said data segments to obtain a plurality of component signals comprises:

Acquiring energy distribution information, local extremum information and trip point information of a frequency spectrum corresponding to the data fragment;

carrying out self-adaptive segmentation on the frequency spectrum corresponding to the data fragment according to the energy distribution information, the local extremum information and the trip point information to obtain a plurality of frequency bands;

constructing a wavelet basis function corresponding to each frequency band;

and decomposing the frequency band according to each wavelet basis function based on an empirical wavelet transformation method to obtain a plurality of component signals.

3. The method of claim 1, wherein said calculating a complexity index for each of said component signals comprises:

constructing a distance matrix corresponding to each component signal;

Performing recursive replacement on the distance matrix to obtain a plurality of replacement matrices corresponding to the distance matrix;

calculating the permutation entropy corresponding to each permutation matrix;

The complexity index of the component signal is obtained according to a plurality of the permutation entropies.

4. The method of claim 1, wherein extracting semantic features from the real-time measurement data, semantically matching and reasoning the semantic features with the device fault ontology library, comprises:

Extracting the semantic features from the real-time measurement data according to a preset semantic feature mapping rule;

Matching the semantic features with the fault information in the equipment fault ontology library to obtain a symptom layer matching result;

And reasoning according to the symptom layer matching result, acquiring a fault mode corresponding to the equipment to be measured, and completing semantic matching and reasoning of the semantic features and the equipment fault ontology library.

5. The method of claim 1, wherein the sub-diagnostic result comprises a plurality of semantic matching results, the outlier component signal corresponding to the semantic matching results, and the raw measurement data corresponding to the semantic matching results; and performing fusion analysis on a plurality of sub-diagnosis results in the preliminary diagnosis results to obtain a final diagnosis conclusion, wherein the fusion analysis comprises the following steps:

Calculating the matching degree of the abnormal component signal corresponding to the semantic matching result and the real-time measurement data;

Screening a plurality of semantic matching results according to the matching degree to obtain the similarity between the screened matching results and the original measurement data;

and determining target similarity among the similarities, wherein the semantic matching result corresponding to the target similarity is the final diagnosis conclusion.

6. The method of claim 1, wherein the measurement strategy comprises a sampling frequency, a site location, and a measurement range; said generating a measurement strategy based on said final diagnostic conclusion and said abnormal component signal, comprising:

Determining the position and the measuring range of the measuring point according to the final diagnosis conclusion;

and determining the sampling frequency according to the frequency range of the abnormal component signal.

7. The method according to claim 1, further comprising, after said generating a measurement strategy to determine a measurement method for said device to be measured based on said final diagnostic conclusion and said abnormal component signal:

Determining a fault early warning threshold corresponding to the equipment to be measured according to the measurement strategy and the final diagnosis conclusion;

And when the measurement parameter corresponding to the real-time measurement data is larger than the fault early warning threshold value, generating an early warning signal to finish early warning and real-time monitoring of the equipment to be measured.

8. The method of claim 7, further comprising, after the generating the pre-warning signal completes the pre-warning and real-time monitoring of the device to be measured:

determining an equipment maintenance strategy of the equipment to be measured according to the fault early warning threshold, the measurement strategy and the final diagnosis conclusion; the maintenance strategy comprises a maintenance period, a maintenance mode and maintenance resource allocation;

Acquiring health information, fault risk information and early warning grade information of the equipment to be measured;

and optimizing the maintenance period, the maintenance mode and the maintenance resource allocation according to the health information, the fault risk information and the early warning level information.

9. A computer device, comprising: a processor and a memory for storing a computer program which when executed by the processor performs the steps of the smart park oriented device measurement method of any one of claims 1 to 8.