CN117851892B

CN117851892B - Intelligent heat supply system data processing method, device and system

Info

Publication number: CN117851892B
Application number: CN202410038700.3A
Authority: CN
Inventors: 李昔真; 王飞飞; 赵永芳; 李晓琴; 刘大海; 东国云
Original assignee: Beijing Zhongneng North Technology Co ltd
Current assignee: Beijing Zhongneng North Technology Co ltd
Priority date: 2024-01-10
Filing date: 2024-01-10
Publication date: 2024-10-01
Anticipated expiration: 2044-01-10
Also published as: CN117851892A

Abstract

According to the intelligent heating system data processing method, device and system provided by the embodiment of the application, the contribution level of the intelligent heating sensing data set to the global fault identification view of the target heating IOT sensing data set is represented by the confidence coefficient for carrying out attention processing on each heating fault representation vector, so that each heating fault attention vector obtained through attention processing can comprehensively and abundantly reflect the detail semantic features of the target heating IOT sensing data set, the global fault identification view obtained according to the fault identification of each heating fault attention vector is as accurate and reasonable as possible, and the fault identification reliability of the heating IOT sensing data set is improved.

Description

Intelligent heat supply system data processing method, device and system

Technical Field

The application relates to the technical field of data processing, in particular to a data processing method, device and system of an intelligent heating system.

Background

With the development of internet of things (IoT) technology, the application of the technology in various fields is becoming wider and wider. In heating systems, ioT sensors are widely used to collect various parameter data, such as temperature, pressure, etc. However, these large amounts of data are not directly usable for fault identification and prediction, and require efficient processing and analysis.

In the past, monitoring heating systems has relied primarily on manual inspection, which is inefficient and may ignore some potential problems. However, the conventional data processing method, such as linear regression or logistic regression, can solve the problem to a certain extent, but still has the problems of low accuracy, inability to process large-scale data, inability to capture complex relationships, and the like.

Therefore, how to effectively process and analyze large-scale heat supply IoT sensing data by using advanced data processing technology to realize accurate and efficient fault identification is a technical problem to be solved currently.

Disclosure of Invention

In order to solve the above problems, the present application provides a method, apparatus and system for processing data in an intelligent heating system.

In a first aspect, there is provided a data processing method of an intelligent heating system, applied to a data processing system, the method comprising:

Disassembling a target heat supply IOT sensing data set to obtain a plurality of intelligent heat supply sensing data sets included in the target heat supply IOT sensing data set;

Performing fault representation vector mining on each intelligent heat supply sensing data set to obtain heat supply fault representation vectors of each intelligent heat supply sensing data set;

according to the heat supply fault representation vector of each intelligent heat supply sensing data set, performing first fault identification processing on each intelligent heat supply sensing data set to obtain a local fault identification view of each intelligent heat supply sensing data set;

for each intelligent heat supply sensing data set, determining a confidence coefficient of the intelligent heat supply sensing data set according to a local fault identification view of the intelligent heat supply sensing data set, and determining a heat supply fault attention vector of the intelligent heat supply sensing data set according to the heat supply fault characterization vector of the intelligent heat supply sensing data set and the confidence coefficient, wherein the confidence coefficient is used for characterizing a contribution level of the intelligent heat supply sensing data set to a global fault identification view of the target heat supply IOT sensing data set;

and performing second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a global fault identification view of the target heat supply IOT sensing data set.

In some aspects, the disassembling the target heat supply IOT sensing dataset to obtain a plurality of intelligent heat supply sensing datasets included in the target heat supply IOT sensing dataset includes:

Determining a data text mask box with a target scanning box size corresponding to the target heat supply IOT sensing data set and a mask interval of the data text mask box;

And utilizing the mask interval to activate the data text mask box, and disassembling the target heat supply IOT sensing data set to obtain a plurality of intelligent heat supply sensing data sets included in the target heat supply IOT sensing data set.

In some aspects, the performing fault-characterization vector mining on each of the intelligent heat-supply sensing data sets to obtain a heat-supply fault-characterization vector of each of the intelligent heat-supply sensing data sets includes:

performing fault characterization vector mining of a first characteristic scale on each intelligent heat supply sensing data set to obtain to-be-processed heat supply fault characterization vectors of the first characteristic scale of each intelligent heat supply sensing data set;

Aggregating the heat supply fault characterization vectors to be processed of the first feature scale of each of the plurality of intelligent heat supply sensing data sets to obtain a heat supply fault characterization vector spectrum to be processed of the target heat supply IOT sensing data set;

And performing fault characterization vector mining of a second feature scale on the to-be-processed heat supply fault characterization vector spectrum to obtain a heat supply fault characterization vector spectrum of the target heat supply IOT sensing data set, wherein the heat supply fault characterization vector spectrum comprises heat supply fault characterization vectors of second feature scales of the intelligent heat supply sensing data sets, and the second feature scales are smaller than the first feature scales.

In some aspects, the determining the confidence coefficient of the intelligent heating sensing dataset according to the local fault recognition perspective of the intelligent heating sensing dataset comprises:

acquiring the viewpoint label number of the first fault identification process, and determining a statistical graph of the viewpoint label number;

Determining a distinction between a local fault recognition perspective of the intelligent heating sensing dataset and the statistical map;

the difference is used as a confidence coefficient of the intelligent heat supply sensing data set.

In some aspects, the determining a heating fault attention vector for the intelligent heating sensing dataset based on the heating fault characterization vector and the confidence coefficient for the intelligent heating sensing dataset comprises:

And weighting the heat supply fault representation vector of the intelligent heat supply sensing data set and the confidence coefficient of the intelligent heat supply sensing data set to obtain the heat supply fault attention vector of the intelligent heat supply sensing data set.

In some aspects, the performing a second fault identification process on the target heat supply IOT sensing dataset according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing datasets to obtain a global fault identification view of the target heat supply IOT sensing dataset includes:

Vector integration operation is carried out on the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets, so that a heat supply fault integration characterization vector is obtained;

and carrying out second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault integration characterization vector to obtain a global fault identification view of the target heat supply IOT sensing data set.

In some aspects, the first fault identification process is implemented by a fault event identification branch of a heating system fault identification network, and the second fault identification process is implemented by a fault scenario identification branch of the heating system fault identification network;

The method further comprises the steps of:

Acquiring a heat supply IOT sensing data set example cluster for debugging the heat supply system fault identification network, wherein the heat supply IOT sensing data set example cluster comprises a plurality of heat supply IOT sensing data set examples, the heat supply IOT sensing data set examples carry first training annotations, the heat supply IOT sensing data set examples comprise a plurality of intelligent heat supply sensing data set examples, and the intelligent heat supply sensing data set examples carry second training annotations;

Through the fault event identification branch, performing first fault identification processing on each intelligent heat supply sensing data set example in each heat supply IOT sensing data set example respectively to obtain fault event identification information of each intelligent heat supply sensing data set example;

for each heating IOT sensing dataset example, the following steps are respectively implemented:

Combining second training notes of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example;

Determining a network training error of the heating system fault identification network according to the difference between the fault event identification information of each intelligent heating sensing data set example and the second training annotation and the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation;

And optimizing network variables of the heating system fault identification network according to the network training errors so as to debug the heating system fault identification network.

In some aspects, the determining the network training error of the heating system fault identification network according to the difference between the fault event identification information and the second training annotation for each of the intelligent heating sensing dataset examples and the difference between the fault identification perspective example and the first training annotation for the heating IOT sensing dataset examples includes:

Determining a unit fault event discrimination error of the heating system fault identification network according to the fault event identification information of each intelligent heating sensing data set example and the difference of the second training annotation;

summing the discrimination errors of each unit fault event of the heat supply system fault identification network to obtain the discrimination errors of the fault event of the heat supply system fault identification network;

determining a fault scene discrimination error of the heating system fault identification network according to the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation;

And carrying out error arrangement on the fault event discrimination errors and the fault scene discrimination errors according to the first error factors of the fault event discrimination errors and the second error factors of the fault scene discrimination errors to obtain network training errors of the heating system fault identification network.

In some aspects, the performing, by the fault event identification branch, a first fault identification process on each of the smart heat supply sensing data set examples in each of the heat supply IOT sensing data set examples to obtain fault event identification information of each of the smart heat supply sensing data set examples includes:

acquiring heat supply fault representation vector examples of each intelligent heat supply sensing data set example in each heat supply IOT sensing data set example;

According to the heat supply fault representation vector examples of each intelligent heat supply sensing data set example, performing first fault identification processing on each intelligent heat supply sensing data set example through the fault event identification branch to obtain fault event identification information of each intelligent heat supply sensing data set example;

The second training annotation combining each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example performs second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example, and the method comprises the following steps:

Determining a data set confidence coefficient of each intelligent heat supply sensing data set example according to a second training annotation of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and determining a heat supply fault strengthening characterization vector example of each intelligent heat supply sensing data set example according to a heat supply fault characterization vector example and the data set confidence coefficient of each intelligent heat supply sensing data set example;

And carrying out second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch according to the heat supply fault strengthening characterization vector examples of a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example.

In some aspects, the determining the data set confidence coefficients for each of the smart heat sensing data set examples based on the second training annotations for each of the smart heat sensing data set examples in the heat IOT sensing data set example comprises:

For each intelligent heat supply sensing data set example in the heat supply IOT sensing data set examples, respectively implementing the following steps:

determining a distinction between the second training annotation of the smart heating sensor dataset example and the statistical map;

the differences are taken as data set confidence coefficients for the smart heating sensor data set example.

In some aspects, before implementing the following steps separately for each heating IOT sensing dataset example, the method further includes:

Determining representative intelligent heat supply sensing data sets of fault keywords obtained by the first fault identification process from a plurality of intelligent heat supply sensing data set examples of the heat supply IOT sensing data set example cluster according to fault event identification information of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example cluster;

For each fault keyword, acquiring an initial characterization vector and a corresponding initial training annotation of the fault keyword, and dynamically optimizing the initial characterization vector by adopting a representative intelligent heat supply sensing data set of the fault keyword to obtain a derivative characterization vector of the fault keyword;

For each intelligent heat supply sensing data set example, determining a commonality score between a heat supply fault representation vector of the intelligent heat supply sensing data set example and a derivative representation vector of each fault keyword, and taking an initial training annotation corresponding to a target derivative representation vector with the highest commonality score as an initial data set training annotation of the intelligent heat supply sensing data set example;

And aiming at each intelligent heat supply sensing data set example, adopting initial data set training annotation of the intelligent heat supply sensing data set example, and dynamically optimizing second training annotation of the intelligent heat supply sensing data set example to obtain derivative training annotation of the intelligent heat supply sensing data set example.

In some aspects, the performing, by the fault scenario recognition branch, a second fault recognition process on the heating IOT sensing dataset example in combination with the second training comments of each of the intelligent heating sensing dataset examples in the heating IOT sensing dataset example to obtain a fault recognition perspective example of the heating IOT sensing dataset example includes: combining the derived training notes of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example;

The determining a network training error of the heating system fault identification network according to the difference between the fault event identification information and the second training annotation of each intelligent heating sensing data set example and the difference between the fault identification point of view example and the first training annotation of the heating IOT sensing data set example comprises: and determining a network training error of the heating system fault identification network according to the difference between the fault event identification information of each intelligent heating sensing data set example and the derivative training annotation and the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation.

In some aspects, the fault event identification information includes a discrimination likelihood that the smart heating sensor data set instance points to each of the fault keywords; the determining, according to the fault event identification information of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example cluster, a representative intelligent heat supply sensing data set of each fault keyword obtained by the first fault identification process from a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example cluster includes:

for each fault keyword, the following steps are respectively implemented:

Determining target intelligent heat supply sensing data set examples which point to the fault keywords and have the possibility of judging from high to low target numbers from a plurality of intelligent heat supply sensing data set examples of the heat supply IOT sensing data set example cluster;

and taking the target intelligent heat supply sensing data set of the target number as a representative intelligent heat supply sensing data set of the target number of the fault keywords.

In some aspects, the number of the representative intelligent heat supply sensing data sets is X, where X is an integer greater than 1, and dynamically optimizing the initial characterization vector to obtain a derivative characterization vector of the fault keyword by using the representative intelligent heat supply sensing data set of the fault keyword includes:

Dynamically optimizing the initial characterization vector by adopting the 1 st representative intelligent heat supply sensing data set of the fault keyword to obtain a1 st initial characterization vector to be processed of the fault keyword;

Dynamically optimizing the (u-1) th to-be-processed initial characterization vector of the fault keyword by adopting the (u) th representative intelligent heat supply sensing data set of the fault keyword to obtain the (u) th to-be-processed initial characterization vector of the fault keyword, wherein u is greater than 0 and not greater than X;

And performing self-adding 1 circulation on the u to obtain an X-th to-be-processed initial characterization vector of the fault keyword, and taking the X-th to-be-processed initial characterization vector of the fault keyword as a derivative characterization vector of the fault keyword.

In some aspects, the number of the representative intelligent heat supply sensing data sets is a plurality, the representative intelligent heat supply sensing data sets using the fault keywords dynamically optimizes the initial characterization vector to obtain derivative characterization vectors of the fault keywords, including:

Obtaining representative heat supply fault representation vectors of each representative intelligent heat supply sensing data set of the fault keywords, and determining heat supply fault average representation vectors of a plurality of representative heat supply fault representation vectors;

Acquiring a first characterization vector confidence coefficient of the heat supply fault average characterization vector and a second characterization vector confidence coefficient of the initial characterization vector;

And performing error arrangement on the heat supply fault average characterization vector and the initial characterization vector according to the first characterization vector confidence and the second characterization vector confidence to obtain the derivative characterization vector of the fault keyword.

In some aspects, the dynamically optimizing the second training annotation of the smart heating sensing dataset with the initial dataset training annotation of the smart heating sensing dataset example to obtain the derivative training annotation of the smart heating sensing dataset example comprises:

Acquiring a first training annotation confidence coefficient of the training annotation of the initial data set and a second training annotation confidence coefficient of the second training annotation;

And performing error arrangement on the training notes of the initial data set and the second training notes according to the first training note confidence level and the second training note confidence level to obtain derivative training notes of the intelligent heat supply sensing data set example.

In a second aspect, there is provided a data processing apparatus comprising:

the data disassembly module is used for disassembling the target heat supply IOT sensing data set to obtain a plurality of intelligent heat supply sensing data sets included in the target heat supply IOT sensing data set;

The vector mining module is used for carrying out fault representation vector mining on each intelligent heat supply sensing data set to obtain heat supply fault representation vectors of each intelligent heat supply sensing data set;

The first recognition module is used for respectively carrying out first fault recognition processing on each intelligent heat supply sensing data set according to the heat supply fault representation vector of each intelligent heat supply sensing data set to obtain a local fault recognition view of each intelligent heat supply sensing data set;

The attention module is used for determining a confidence coefficient of the intelligent heat supply sensing data set according to the local fault recognition view of the intelligent heat supply sensing data set and determining a heat supply fault attention vector of the intelligent heat supply sensing data set according to the heat supply fault characterization vector of the intelligent heat supply sensing data set and the confidence coefficient, wherein the confidence coefficient is used for representing the contribution level of the intelligent heat supply sensing data set to the global fault recognition view of the target heat supply IOT sensing data set;

and the second recognition module is used for carrying out second fault recognition processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a global fault recognition view of the target heat supply IOT sensing data set.

In a third aspect, a data processing system is provided, comprising a processor and a memory in communication with each other, the processor being arranged to retrieve a computer program from the memory and to implement the method of the first aspect by running the computer program.

In a fourth aspect, a computer readable storage medium is provided, on which a computer program is stored, which computer program, when run, implements the method of the first aspect.

When the fault identification is carried out on the target heat supply IOT sensing dataset, the method, the device and the system for processing the intelligent heat supply system data provided by the embodiment of the application firstly disassemble the target heat supply IOT sensing dataset to obtain a plurality of intelligent heat supply sensing datasets which are included in the target heat supply IOT sensing dataset, then respectively carry out fault characterization vector mining on each intelligent heat supply sensing dataset to obtain heat supply fault characterization vectors of each intelligent heat supply sensing dataset, and respectively carry out first fault identification processing on each intelligent heat supply sensing dataset according to the heat supply fault characterization vectors of each intelligent heat supply sensing dataset to obtain a local fault identification view of each intelligent heat supply sensing dataset; thereby determining the confidence coefficient of each intelligent heat supply sensing data set according to the local fault recognition viewpoint of each intelligent heat supply sensing data set, and performing attention processing on each heat supply fault characterization vector according to the heat supply fault characterization vector and the confidence coefficient of each intelligent heat supply sensing data set, and finally, performing second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the intelligent heat supply sensing data sets to obtain the global fault identification view of the target heat supply IOT sensing data set. The confidence coefficient for carrying out attention processing on each heat supply fault characterization vector characterizes the contribution level of the intelligent heat supply sensing data set to the global fault recognition view of the target heat supply IOT sensing data set, so that each heat supply fault attention vector obtained through attention processing can comprehensively and abundantly reflect the detail semantic features of the target heat supply IOT sensing data set, the global fault recognition view obtained according to fault recognition of each heat supply fault attention vector is accurate and reasonable as much as possible, and the fault recognition reliability of the heat supply IOT sensing data set is improved.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the embodiments will be briefly described below, it being understood that the following drawings only illustrate some embodiments of the present application and therefore should not be considered as limiting the scope, and other related drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a flowchart of a data processing method of an intelligent heating system according to an embodiment of the application.

Detailed Description

In order to better understand the above technical solutions, the following detailed description of the technical solutions of the present application is made by using the accompanying drawings and specific embodiments, and it should be understood that the specific features of the embodiments and the embodiments of the present application are detailed descriptions of the technical solutions of the present application, and not limiting the technical solutions of the present application, and the technical features of the embodiments and the embodiments of the present application may be combined with each other without conflict.

Fig. 1 shows a smart heating system data processing method, applied to a data processing system, comprising the following steps 110-150.

Step 110, disassembling the target heat supply IOT sensing dataset to obtain a plurality of intelligent heat supply sensing datasets included in the target heat supply IOT sensing dataset.

And 120, performing fault representation vector mining on each intelligent heat supply sensing data set to obtain heat supply fault representation vectors of each intelligent heat supply sensing data set.

And 130, respectively performing first fault identification processing on each intelligent heat supply sensing data set according to the heat supply fault representation vector of each intelligent heat supply sensing data set to obtain a local fault identification view of each intelligent heat supply sensing data set.

Step 140, determining, for each intelligent heat supply sensing dataset, a confidence coefficient of the intelligent heat supply sensing dataset according to the local fault recognition perspective of the intelligent heat supply sensing dataset, and determining a heat supply fault attention vector of the intelligent heat supply sensing dataset according to the heat supply fault characterization vector and the confidence coefficient of the intelligent heat supply sensing dataset, wherein the confidence coefficient is used for characterizing a contribution level of the intelligent heat supply sensing dataset to the global fault recognition perspective of the target heat supply IOT sensing dataset.

And 150, performing second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a global fault identification view of the target heat supply IOT sensing data set.

When the embodiment of the application is applied, when the target heat supply IOT sensing dataset is subjected to fault identification, the target heat supply IOT sensing dataset is disassembled to obtain a plurality of intelligent heat supply sensing datasets contained in the target heat supply IOT sensing dataset, then fault characterization vector mining is respectively carried out on each intelligent heat supply sensing dataset to obtain heat supply fault characterization vectors of each intelligent heat supply sensing dataset, and first fault identification processing is respectively carried out on each intelligent heat supply sensing dataset according to the heat supply fault characterization vectors of each intelligent heat supply sensing dataset to obtain a local fault identification view of each intelligent heat supply sensing dataset; thereby determining the confidence coefficient of each intelligent heat supply sensing data set according to the local fault recognition viewpoint of each intelligent heat supply sensing data set, and performing attention processing on each heat supply fault characterization vector according to the heat supply fault characterization vector and the confidence coefficient of each intelligent heat supply sensing data set, and finally, performing second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the intelligent heat supply sensing data sets to obtain the global fault identification view of the target heat supply IOT sensing data set. The confidence coefficient for carrying out attention processing on each heat supply fault characterization vector characterizes the contribution level of the intelligent heat supply sensing data set to the global fault recognition view of the target heat supply IOT sensing data set, so that each heat supply fault attention vector obtained through attention processing can comprehensively and abundantly reflect the detail semantic features of the target heat supply IOT sensing data set, the global fault recognition view obtained according to fault recognition of each heat supply fault attention vector is accurate and reasonable as much as possible, and the fault recognition reliability of the heat supply IOT sensing data set is improved.

In step 110, the target heating IOT sensing dataset is a dataset collected by a plurality of sensors deployed throughout the heating system, including critical parts of boilers, piping, radiators, etc. The data set may contain various types of data, such as temperature, pressure, flow, etc., all collected in real time through internet of things. The intelligent heating sensing dataset is a subset of the total IOT sensing dataset that is functional or representative of a particular system component. For example, one data set may contain only the data of the boiler and the other data set may contain the data of the pipeline.

In step 110, the target heating IOT sensing dataset contains data of the temperature, pressure, flow rate of the pipes, heat output of the radiator, etc. of the boiler, while the intelligent heating sensing dataset may be processed and analyzed only for the data of the boiler. Each intelligent heating sensing data set is processed separately and a respective fault characterization vector is mined. Next, the technical scheme of step 110 is illustrated. In a heating system, there are three main components: boiler, pipeline and radiator. These three portions of data constitute the target heating IOT sensing dataset. In step 110, the entire data set is disassembled into three intelligent heating sensing data sets, each data set containing its own sensor data.

Fault-characterization vector mining is one method of extracting valuable information from the intelligent heat supply sensing dataset in step 120. It generates feature vectors representing device or system fault conditions by identifying and analyzing data patterns. These vectors may be used to help monitor system conditions, predict faults, and optimize maintenance planning.

For example, a heating system includes components such as a water pump, piping, and a radiator, each of which is equipped with sensors to collect operational data. If the current of the water pump suddenly rises, while the temperature and pressure remain unchanged, this may indicate that the water pump is problematic. In this case, the fault-characterization vector may be [1, 0], indicating that the water pump is faulty, the piping is normal, and the radiator is normal. The heat supply fault characterization vector is a unified description of single or multiple fault characterization vectors, which can reflect the health status of the entire heat supply system.

For example, in the above example, if there are three pumps, each with its own fault-characterizing vector, then the heating fault-characterizing vector may be a combination of the three vectors or some form of integration, such as [1,0,0,0,1,0,0,0,1], which indicates that the first and second pumps are problematic, while the third pump is normal.

The technical solution of step 120 may be illustrated by the following example: assume that there is a set of heating system data comprising three water pumps. Firstly, carrying out fault representation vector mining on the data of each water pump to obtain fault representation vectors of each water pump. A heating fault characterization vector is then generated from these vectors to describe the state of the overall heating system. In this way, when a problem occurs with the system, it is possible to quickly locate which part in particular has the problem, so that maintenance and optimization can be performed more efficiently.

In some examples, the heating fault characterization vector is a vector that represents possible fault patterns or signatures in the dataset. It may include various different types of data such as temperature, pressure, flow, etc., and use this data to determine possible faults.

For example, there is a smart heating system comprising three parts, a boiler, a pipe and a radiator. Each section has its own sensor, collecting the respective data. All of this data is then integrated together to form the target heating IOT sensing dataset. Now, this dataset is disassembled to obtain three intelligent heating sensing data sets: a boiler data set, a duct data set, and a radiator data set. Next, a fault-characterization vector needs to be mined for each data set. Taking the boiler dataset as an example, the fault-characterization vector may include the following elements:

abnormal temperature: if the temperature of the boiler is too high or too low, this may mean that there is a malfunction;

pressure anomaly: also, if the pressure of the boiler exceeds or falls below the normal range, it may be indicative of a problem;

abnormal fuel consumption: if the fuel consumption of the boiler suddenly increases or decreases, it may mean that the efficiency of the boiler is changed, and there may be a malfunction;

abnormal discharge: if the amount of emissions (e.g. carbon dioxide) from the boiler exceeds normal values, it may also be an indication of a malfunction.

These four elements constitute a fault-characterizing vector of the boiler data set. Each element represents one possible failure mode or feature. The same method can be used to mine fault-characterization vectors for other data sets. By the method, faults of a single part can be detected, and the overall fault identification can be carried out on the whole system, so that the problem can be more accurately positioned, and the stability and the efficiency of the heating system are improved.

For another example, there is a simple heating system consisting of three parts: boiler (Boiler), water Pump (Pump) and radiator (Radiator). Each section has sensors to collect operational data. In this example, the following features may be defined:

A boiler: temperature, pressure;

and (3) a water pump: current and vibration;

a radiator: temperature, humidity.

For normal operating conditions, the following range of values may be desired:

a boiler: the temperature is between 50 and 80 ℃ and the pressure is between 1 and 2 MPa;

and (3) a water pump: the current is between 10 and 15A, and the vibration is between 0 and 0.05 g;

a radiator: the temperature is between 20 and 30 ℃ and the humidity is between 30 and 50 percent.

Now, assume that the following readings are received from the sensors:

a boiler: the temperature is 90 ℃ and the pressure is 1.5MPa;

and (3) a water pump: current 14A, vibration 0.07g;

A radiator: the temperature is 25 ℃ and the humidity is 45%.

The data may be converted into a fault-characterization vector. For each portion, if all readings are within the normal range, a 0 is assigned; if either reading is outside the normal range, a 1 is assigned. In this example, the fault-characterization vector would be [1, 0] because the temperature of the boiler and the vibration of the water pump are outside of normal ranges.

In step 130, the first fault identification process refers to a preliminary fault identification process performed on the intelligent heating sensing dataset. By using the heating fault-characterization vector mined in step 120, it is analyzed and determined whether a fault exists in the portion of the device or system represented by the data set. For example, if an abnormal rise in temperature is found in the data set of the boiler while the pressure continues to drop, this may be considered a fault flag and marked as a fault in the first fault identification process.

The local fault recognition perspective is a judgment or perspective of whether a fault exists in the single intelligent heat supply sensing data set according to the result of the first fault recognition processing. In the foregoing example, for example, the conclusion based on fault-characterization vectors (e.g., abnormal temperature and pressure) of the boiler data set is a local fault-recognition perspective. This view tells the boiler that there may be a fault.

The technical solution of step 130 may be described by the following example:

In step 120, three intelligent heating sensing data sets (boiler, piping, radiator) are mined for fault characterization vectors. Next, in step 130, a first failure recognition process is required for the three data sets. The fault identification may be performed by, for example, a machine learning model or other algorithm, based on the heating fault characterization vector for each data set. For example, if the temperature and pressure data of the boiler data set show abnormality, it can be judged that there is a possibility of a failure of the boiler, which is a local failure recognition viewpoint. Similarly, the same processing can also be performed on the channels and the radiator data sets to obtain their respective local fault identification perspectives. By the method, specific problems of the system can be found and positioned in early stage, so that the system can be repaired in time, and the stability and efficiency of the heating system are improved.

In step 140, the confidence coefficient is a measure characterizing the impact or contribution of a certain intelligent heat supply sensing dataset to the global fault identification perspective. If the confidence coefficient of a data set is high, the greater the impact of the data set in global fault identification. The heating fault attention vector combines the heating fault characterization vectors and confidence coefficients of each intelligent heating sensing dataset to generate a vector describing the global state of the system. This vector may help to better understand the global state of the system, in particular the likelihood and location of faults. The global fault recognition view is an evaluation of whether the whole heating system has faults or not and the possible positions of the faults after integrating the information of all intelligent heating sensing data sets and considering the respective confidence coefficients. The contribution level refers to the contribution degree of each intelligent heat supply sensing data set to the global fault identification viewpoint, and the contribution degree is determined by the confidence coefficient.

Next, step 140 is described by a specific example: for example, three intelligent heating sensing data sets: boiler, water pump and radiator. In the preceding step, a fault-characterization vector has been mined for each data set, and preliminary fault-recognition processing is performed.

Now, in step 140, the confidence coefficient for each data set is first determined. Assuming that by some means (e.g., based on historical data or expert knowledge), the confidence coefficient for the boiler dataset is 0.5, the confidence coefficient for the water pump dataset is 0.3, and the confidence coefficient for the radiator dataset is 0.2.

And then multiplying the fault representation vector of each data set by the corresponding confidence coefficient to obtain the heat supply fault attention vector of each data set. For example, if the fault-characterization vector for a boiler dataset is [1, 0], then the heating fault-attention vector for the boiler is [0.5, 0].

And finally, adding the heat supply fault attention vectors of all the data sets to obtain the global fault identification viewpoint of the system. This view may help to more accurately understand the state of the system, particularly the possible faults.

Furthermore, in the deep learning and attention mechanism, the heating fault attention vector should be a multidimensional feature vector that weights each heating fault characterization vector by confidence coefficients. For example, if there are three intelligent heating sensing data sets (boiler, water pump, radiator), each data set has its own fault-characterization vector (e.g., a one-dimensional array of length n) and confidence coefficient. The heat supply fault concentration vector can be obtained by multiplying the fault representation vector of each data set by the corresponding confidence coefficient and then adding the obtained results.

For example, assume that the fault-characterization vectors for each dataset are f1, f2, and f3, respectively, and their corresponding confidence coefficients are c1, c2, and c3, respectively. Then the first time period of the first time period, the attention vector of heat supply faults can be obtained calculated as c1f1+c2f2+c3×f3. In this way a multidimensional feature vector is obtained which is able to represent the state of the entire heating system, i.e. a heating fault concentration vector. The value of each dimension reflects the extent to which a particular feature contributes to a system fault, thereby helping to more accurately understand the state of the system, particularly the possible faults.

In step 150, the second fault identification process refers to a global fault identification process performed on the entire target heat supply IOT sensing dataset using the local fault identification views and their confidence coefficients after the first fault identification process is performed on each intelligent heat supply sensing dataset and the respective local fault identification views are obtained.

For example, the heat supply fault attention vectors (product of local fault identification perspective and confidence coefficient) of the respective data sets may be weighted averaged or otherwise combined to generate a global fault identification perspective. This global fault identification view reflects the health status of the entire heating system, helping to understand and locate possible problems at a global level.

The technical solution of step 150 may be illustrated by the following example: three intelligent heating sensing data sets (boiler, water pump, radiator) have been obtained in the first few steps for heating fault attention vectors: [0.5, 0], [0,0.3,0], [0,0,0.2].

In step 150, a second fault identification process is required. This can be achieved by summing the three heating fault attention vectors, resulting in a global heating fault attention vector: [0.5,0.3,0.2]. A threshold value, such as 0.4, may then be set, and if any element exceeds this threshold value, the corresponding portion is considered likely to be faulty.

In this example, the global fault identification perspective is that the boiler may be faulty because element (0.5) of the boiler exceeds a threshold. In this way, the fault condition of the entire heating system can be more fully understood and recognized, so that targeted maintenance or optimization measures can be taken.

The beneficial effects of the technical scheme are mainly shown in the following aspects:

(1) System level fault identification: and (3) obtaining each intelligent heat supply sensing data set by disassembling the target heat supply IOT sensing data set, and then carrying out fault characterization vector mining and fault identification processing on each data set, so that fault detection and identification of a system level can be realized. The method not only can find out which parts possibly have problems, but also can further locate specific fault modes or characteristics;

(2) Early fault early warning: according to the heat supply fault representation vectors of each intelligent heat supply sensing data set, first fault identification processing is timely carried out to obtain a local fault identification view, and early faults can be found and early warned, so that the faults are prevented from being enlarged or other problems are prevented from being caused;

(3) Fault identification from a global perspective: by combining the local fault recognition perspective and the confidence coefficient of each intelligent heat supply sensing data set to generate a heat supply fault attention vector, fault recognition at the global perspective can be realized. This approach takes into account not only the state of each part, but also their contribution to the global state, and thus may provide a more comprehensive, accurate fault identification result.

(4) The fault diagnosis efficiency is improved: through the second fault identification processing, the global fault identification view of the target heating IOT sensing dataset can be obtained quickly. Thus, possible problems can be found out in a short time, and the efficiency of fault diagnosis is greatly improved.

In general, by utilizing the data of each part of the heating system, the attention mechanism and the fault characterization vector, efficient and accurate fault identification is realized, and the stable operation of the heating system is facilitated.

In some preferred embodiments, the disassembly of the target heating IOT sensing dataset in step 110 results in a number of intelligent heating sensing datasets comprised by the target heating IOT sensing dataset, including steps 111-112.

Step 111, determining a data text mask box with a target scanning frame size corresponding to the target heat supply IOT sensing data set and a mask interval of the data text mask box.

And 112, activating the data text mask box by using the mask interval, and disassembling the target heat supply IOT sensing data set to obtain a plurality of intelligent heat supply sensing data sets included in the target heat supply IOT sensing data set.

Wherein the data text mask box is a tool or method for selecting data of a particular size from a target heating IOT sensing dataset. Similar to a sliding window in computer vision, it can move through a dataset and pick a portion of the data at each location. The target scan box size refers to the size or range of the data text mask box. It determines how much data to choose at a time to process. The mask interval is the step size in which the data text mask box moves in the dataset. It determines how to select data from the dataset, if the mask interval is small, the selected data will overlap significantly; conversely, if the mask interval is large, then the selected data may miss some information.

For example, there is a target heating IOT sensing dataset that contains operational data for an entire day of heating systems. In step 111, a data text mask box having a target scan box size is determined, such as selecting data for one hour at a time. At the same time, a masking interval is also determined, such as selecting to move the data text mask box once every 30 minutes. Then, in step 112, the data text mask box is activated with this mask interval to disassemble the target heat supply IOT sensing dataset. Specifically, the first hour of data in the dataset is selected as the first intelligent heating sensing dataset, then one half hour of data is moved, the next hour of data is selected as the second dataset, and so on, until the end of the dataset.

In this way, a large data set can be broken down into a series of small data sets, each representing a heating system status for a different period of time. These data sets may then be subjected to fault-characterization vector mining, fault-recognition processing, etc., to better understand and monitor the operation of the heating system. The technical scheme can help to more effectively process large-scale heat supply IOT sensing data sets. By reasonably setting the data text mask box and the mask interval, the calculation efficiency can be ensured, and meanwhile, valuable information in the data set can be extracted as much as possible, so that the faults of the heating system can be more accurately identified, and the stability and the efficiency of the heating system are improved.

In some preferred embodiments, the fault-characterizing vector mining of each of the intelligent heat-supplying sensing data sets in step 120 obtains a heat-supplying fault-characterizing vector of each of the intelligent heat-supplying sensing data sets, including steps 121-123.

And 121, performing fault representation vector mining of a first characteristic scale on each intelligent heat supply sensing data set to obtain to-be-processed heat supply fault representation vectors of the first characteristic scale of each intelligent heat supply sensing data set.

And 122, aggregating the to-be-processed heat supply fault characterization vectors of the first feature scales of the plurality of intelligent heat supply sensing data sets to obtain a to-be-processed heat supply fault characterization vector spectrum of the target heat supply IOT sensing data set.

Step 123, performing fault representation vector mining of a second feature scale on the to-be-processed heat supply fault representation vector spectrum to obtain a heat supply fault representation vector spectrum of the target heat supply IOT sensing dataset, wherein the heat supply fault representation vector spectrum comprises heat supply fault representation vectors of second feature scales of the intelligent heat supply sensing datasets, and the second feature scales are smaller than the first feature scales.

The first feature scale is a feature scale used when the fault characterization vector of each intelligent heat supply sensing data set is mined, and can be understood as the thickness degree of feature extraction or the visual angle width of data analysis. The heat supply fault representation vector to be processed refers to a fault representation vector which is mined from each intelligent heat supply sensing data set based on the first characteristic scale, but is not subjected to global aggregation. The heat supply fault representation vector spectrum refers to a set of fault representation vectors comprising all intelligent heat supply sensing data sets, which is obtained after the fault representation vectors of the second feature scale are mined. The second feature scale, which is a feature scale used when further mining the heat supply fault representation vector spectrum to be processed, is smaller than the first feature scale, meaning that the fault features will be analyzed from a finer perspective, deeper.

For example, there are three intelligent heating sensing data sets: boiler, water pump, radiator. In step 121, the fault-characterizing vectors are mined for each data set on the first feature scale, so as to obtain respective heat-supplying fault-characterizing vectors to be processed. Then, in step 122, the three heat supply fault representation vectors to be processed are aggregated to obtain a heat supply fault representation vector spectrum to be processed including global information. Next, in step 123, this heat supply fault representation vector spectrum to be processed is further mined on the second feature scale, so as to obtain a final heat supply fault representation vector spectrum. This characterization vector spectrum contains finer, more extensive fault signature information.

Thus, by mining fault characterization vectors on two different feature scales, deep fault identification from local to global is achieved. On a first feature scale, a local problem for each device or part can be found; on the second feature scale, fault features that may be ignored can again be found from a global perspective. The method improves the accuracy and the robustness of fault identification, and is helpful for finding out potential heating system problems earlier and more accurately.

In some preferred embodiments, the determining of the confidence coefficient of the intelligent heating sensing dataset according to the perspective of the local fault identification of the intelligent heating sensing dataset in step 140 comprises steps 141-143.

Step 141, obtaining the number of viewpoint labels of the first fault identification process, and determining a statistical graph of the number of viewpoint labels.

Step 142, determining the distinction between the local fault recognition perspective of the intelligent heating sensing dataset and the statistical map.

Step 143, using the difference as a confidence coefficient of the intelligent heat supply sensing data set.

The viewpoint label number refers to the label number corresponding to each local failure recognition viewpoint in the first failure recognition process. That is, the number of times each possible failure mode or state is identified may be counted. A statistical chart is a graph or data structure generated based on the number of perspective labels to show the frequency distribution of each possible failure mode or state. The difference refers to the difference between the local fault recognition perspective and the statistical map of the intelligent heating sensing data set. If the local failure recognition perspective of a data set differs significantly from the overall failure mode or state distribution, then its confidence coefficient will be higher.

For example, in the first few steps a target heating IOT sensing dataset comprising three intelligent heating sensing datasets (boiler, water pump, radiator) has been disassembled and respective local fault recognition perspectives obtained.

In step 141, the number of perspective labels of the first failure recognition process is first obtained, and a statistical map is generated. For example, it is found that the failure mode a of the boiler is recognized 10 times, the failure mode B is recognized 5 times, and the no-failure state is recognized 85 times. Then, in step 142, a distinction between the local fault identification perspective and the statistical map for each data set is determined. For example, if the local failure recognition perspective of the boiler is failure mode a, it is distinguished from the statistical chart by 10/100=0.1. Finally, in step 143, this distinction is taken as a confidence coefficient for the boiler data set. This means that the contribution level of the boiler dataset to the global fault identification perspective is 0.1.

The above embodiments provide a way to determine the confidence coefficient of each intelligent heating sensing dataset based on statistical analysis, so that the contribution level of each dataset to the global fault identification perspective can be more accurately assessed. In addition, this approach can help discover those data sets that deviate from normal for further analysis and processing.

In some preferred embodiments, determining a heating fault concentration vector for the intelligent heating sensing dataset from the heating fault characterization vector and the confidence coefficient of the intelligent heating sensing dataset in step 140 comprises: and weighting the heat supply fault representation vector of the intelligent heat supply sensing data set and the confidence coefficient of the intelligent heat supply sensing data set to obtain the heat supply fault attention vector of the intelligent heat supply sensing data set.

For example, there are three intelligent heating sensing data sets: boiler, water pump, radiator. In the previous step, a heating fault-indicative vector has been obtained for each data set, such as f1, f2, f3, respectively, and corresponding confidence coefficients, c1, c2, c3, respectively.

In this step, it is necessary to determine the heating fault concentration vector for each data set based on the heating fault characterization vector and the confidence coefficient. The specific operation is to weight the heating fault characterization vector and the confidence coefficient. For example, for a boiler, its heat supply fault attention vector can be obtained by multiplying f1 by c 1; likewise, the heat supply failure attention vectors of the water pump and the radiator can also be calculated in the same manner.

It can be seen that by weighting the heating fault characterization vector and the confidence coefficient, the calculation of the heating fault attention vector for each intelligent heating sensing dataset is achieved. The method fully considers the fault characteristics of each data set and the contribution degree of each data set to the global fault identification viewpoint, so that the potential heating system problem can be identified more accurately. Meanwhile, the method also improves the flexibility and adaptability of fault identification, and the weights of different data sets can be adjusted according to actual conditions so as to meet different requirements.

In some preferred embodiments, the performing the second fault identification process on the target heat supply IOT sensing dataset according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing datasets in step 150 to obtain a global fault identification view of the target heat supply IOT sensing dataset includes steps 151-152.

And 151, performing vector integration operation on the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a heat supply fault integration characterization vector.

And 152, performing a second fault identification process on the target heat supply IOT sensing data set according to the heat supply fault integrated characterization vector, so as to obtain a global fault identification view of the target heat supply IOT sensing data set.

The vector integration operation is to combine a plurality of vectors (in this case, the heat supply fault attention vectors of each intelligent heat supply sensing data set) in some way to obtain a new vector. The integration operation may be a simple weighted average or may be a more complex operation such as Principal Component Analysis (PCA) or other dimension reduction technique. The heat supply fault integrated characterization vector is a new vector obtained after vector integration operation, integrates the information of all heat supply fault attention vectors, and can reflect the state of a heat supply system from the global angle.

For example, there are three intelligent heating sensing data sets (boiler, water pump, radiator) heating fault attention vectors: f1 = [0.5, 0], f2= [0,0.3,0], f3= [0,0,0.2].

In step 151, a vector integration operation is first performed. Assuming that the selected integration operation is a weighted average, these three vectors may be added and then divided by 3 to yield a heating fault integration characterization vector f= [0.17,0.1,0.07]. Then, in step 152, a second fault identification process is performed on the target heating IOT sensing dataset based on this heating fault integration characterization vector. For example, a threshold may be set, and if any element exceeds this threshold, the corresponding portion is considered to be likely to be faulty. In this example, the global fault identification perspective is that the system does not find a significant fault, since all elements are below the threshold.

In this way, fault identification from multiple local perspectives to one global perspective is achieved through vector integration operations. In this way, not only the status of each section can be taken into account, but also the health of the heating system as a whole can be understood, thereby more accurately identifying and locating potential problems. At the same time, this approach also increases the efficiency of fault identification because only one integrated token vector needs to be analyzed, rather than having to process the data for each part separately.

In some examples, the first fault identification process is implemented by a fault event identification branch of a heating system fault identification network and the second fault identification process is implemented by a fault scenario identification branch of the heating system fault identification network. Based thereon, the method further comprises: acquiring a heat supply IOT sensing data set example cluster for debugging the heat supply system fault identification network, wherein the heat supply IOT sensing data set example cluster comprises a plurality of heat supply IOT sensing data set examples, the heat supply IOT sensing data set examples carry first training annotations, the heat supply IOT sensing data set examples comprise a plurality of intelligent heat supply sensing data set examples, and the intelligent heat supply sensing data set examples carry second training annotations; through the fault event identification branch, performing first fault identification processing on each intelligent heat supply sensing data set example in each heat supply IOT sensing data set example respectively to obtain fault event identification information of each intelligent heat supply sensing data set example; for each heating IOT sensing dataset example, the following steps are respectively implemented: combining second training notes of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example; determining a network training error of the heating system fault identification network according to the difference between the fault event identification information of each intelligent heating sensing data set example and the second training annotation and the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation; and optimizing network variables of the heating system fault identification network according to the network training errors so as to debug the heating system fault identification network.

Wherein, the heating system fault recognition network is a neural network model specially designed for the recognition of the heating system faults. It may contain multiple branches for handling different tasks or fault identification from different angles. The fault event identification branch is a component of a heat supply system fault identification network and is mainly responsible for performing a first fault identification process, namely identifying a local fault of each intelligent heat supply sensing data set. The fault scene recognition branch is also a component part of a heat supply system fault recognition network and is mainly responsible for carrying out second fault recognition processing, namely carrying out global fault recognition on the whole heat supply system according to the local fault recognition viewpoints and confidence coefficients of each intelligent heat supply sensing data set. A heat supply IOT sensing dataset example cluster is a dataset for training and debugging a heat supply system fault identification network, wherein a plurality of heat supply IOT sensing dataset examples are included. The first training annotation is an annotation of the global fault status for each heating IOT sensing dataset instance. The second training annotation is an annotation of the local fault status for each example intelligent heating sensing dataset. The network training error is an index for measuring the difference between the predicted result and the actual result in the training process of the model. It is often desirable to minimize this error by optimizing the network variables.

For example, in a large heating system, various internet of things devices are installed, and operation data of the heating system is collected. These data are then consolidated into a set (cluster) of samples, including temperature, pressure, flow, etc., and the fault event and fault scenario labels are provided by the expert. And then training a heat supply system fault recognition network by using the samples, and processing the samples through a fault event recognition branch and a fault scene recognition branch to obtain a fault recognition result. According to the difference between the prediction result and the actual label, calculating a network training error, and optimizing parameters of the network according to the error, thereby improving the accuracy of fault identification.

Therefore, a large amount of data collected by the Internet of things equipment can be effectively utilized, faults in the heating system can be automatically identified through a machine learning method, the speed and accuracy of fault identification can be remarkably improved, detailed fault information can be provided for maintenance personnel, the maintenance personnel can be helped to locate and repair the faults more quickly, and accordingly the running efficiency and stability of the heating system are improved.

In some alternative embodiments, the determining the network training error of the heating system fault identification network based on the difference between the fault event identification information and the second training annotation for each of the intelligent heating sensing dataset examples and the difference between the fault identification perspective example and the first training annotation for the heating IOT sensing dataset examples comprises: determining a unit fault event discrimination error of the heating system fault identification network according to the fault event identification information of each intelligent heating sensing data set example and the difference of the second training annotation; summing the discrimination errors of each unit fault event of the heat supply system fault identification network to obtain the discrimination errors of the fault event of the heat supply system fault identification network; determining a fault scene discrimination error of the heating system fault identification network according to the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation; and carrying out error arrangement on the fault event discrimination errors and the fault scene discrimination errors according to the first error factors of the fault event discrimination errors and the second error factors of the fault scene discrimination errors to obtain network training errors of the heating system fault identification network.

In the above technical solution, it is mainly introduced how to train a heating system fault recognition network by using intelligent heating sensing data and heating IOT sensing data, and to determine the network training error of the network by calculating and sorting various errors.

The fault event discrimination error and the fault scene discrimination error respectively reflect the recognition capability of the network to the fault event and the fault scene. By comparing the output of the network with known annotations (i.e., actually occurring fault events or scenarios), both errors can be calculated.

The first error factor and the second error factor are used for balancing the importance of the fault event discrimination error and the fault scene discrimination error in the network training error. For example, if the identification of the fault event is considered more important than the identification of the fault scenario, the first error factor may be set to be larger.

The network training error is an integral index and reflects the integral identification capability of the network to the faults of the heating system. By minimizing this error, the performance of the network can be optimized.

For example, there is a heating system that includes a number of sensors and IOT devices. These devices will continue to collect data as the system is running. These data can then be used to train a heating system failure recognition network. In the training process, various errors are calculated, and then weighted summation is carried out according to a preset error factor, so that the network training error is obtained. By optimizing this error, the fault recognition capability of the network can be improved.

Therefore, by introducing intelligent heat supply sensing data and heat supply IOT sensing data and accurate error calculation and balance, the training of the heat supply system fault identification network is more accurate, and the accuracy and efficiency of fault identification are improved.

In some optional embodiments, the performing, by the fault event identification branch, a first fault identification process on each of the smart heat supply sensing data set examples in each of the heat supply IOT sensing data set examples to obtain fault event identification information of each of the smart heat supply sensing data set examples includes: acquiring heat supply fault representation vector examples of each intelligent heat supply sensing data set example in each heat supply IOT sensing data set example; according to the heat supply fault representation vector examples of each intelligent heat supply sensing data set example, performing first fault identification processing on each intelligent heat supply sensing data set example through the fault event identification branch to obtain fault event identification information of each intelligent heat supply sensing data set example; the second training annotation combining each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example performs second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example, and the method comprises the following steps: determining a data set confidence coefficient of each intelligent heat supply sensing data set example according to a second training annotation of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and determining a heat supply fault strengthening characterization vector example of each intelligent heat supply sensing data set example according to a heat supply fault characterization vector example and the data set confidence coefficient of each intelligent heat supply sensing data set example; and carrying out second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch according to the heat supply fault strengthening characterization vector examples of a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example.

An example of a heating fault characterization vector is a multi-dimensional vector, each dimension representing a feature for describing and representing fault information of a heating system.

The smart heating sensor dataset confidence coefficient is a weight value that represents the degree of confidence or importance of the smart heating sensor dataset example. For example, certain critical, highly sensitive sensor data may be given a higher confidence coefficient.

The heat supply fault strengthening characterization vector example is obtained by combining the confidence coefficient of the intelligent heat supply sensing data set on the basis of the heat supply fault characterization vector example, and the weighted multidimensional vector is used for representing fault information with strengthened importance consideration.

For example, a collection of heating IOT sensing dataset examples is obtained from a large heating system, including various information such as temperature, pressure, flow, etc. Firstly, a heat supply fault characterization vector example is generated for each intelligent heat supply sensing data set example, and then fault event identification processing is carried out according to the characterization vectors through fault event identification branches. Next, a confidence coefficient for each instance of the intelligent heating sensing dataset is determined from the second training annotations, and then this confidence coefficient and the characterization vector are combined to generate a heating fault-intensive characterization vector instance. And finally, inputting the enhanced characterization vectors into a fault scene recognition branch to perform fault scene recognition processing.

The technical scheme introduces the concepts of the confidence coefficient and the reinforced characterization vector, and can more accurately represent and identify the fault information of the heating system. In practice, the confidence coefficient can be adjusted to pay attention to more important or critical sensing data, so that the accuracy of fault identification is improved. Meanwhile, by using the enhanced characterization vector, subtle changes in fault information can be better captured, so that the sensitivity of fault identification is improved. In general, the technical scheme can effectively improve the accuracy and the sensitivity of the fault identification of the heating system, thereby improving the operation efficiency and the stability of the heating system.

In some alternative embodiments, the determining the data set confidence coefficients for each of the smart heat sensing data set examples based on the second training annotations for each of the smart heat sensing data set examples in the heat IOT sensing data set example includes: for each intelligent heat supply sensing data set example in the heat supply IOT sensing data set examples, respectively implementing the following steps: acquiring the viewpoint label number of the first fault identification process, and determining a statistical graph of the viewpoint label number; determining a distinction between the second training annotation of the smart heating sensor dataset example and the statistical map; the differences are taken as data set confidence coefficients for the smart heating sensor data set example.

Wherein the data set confidence coefficient is a measure representing the degree of certainty of the model in its predicted outcome. If the model predicts a result that is very close to the actual tag (i.e., training annotation), then the confidence coefficient will be high; conversely, if the model predicts a large difference from the actual label, the confidence coefficient will be low. The number of point-of-view labels in the first failure recognition process refers to the number of failure events of a specific type in the prediction result obtained by the first failure recognition process. The statistical graph is a visualization tool used for describing information such as data distribution, trend and the like. Here, the statistical map may be a histogram, pie chart, or other type of chart describing the distribution of the point of view tags.

For example, there is a heating system that includes a number of sensors and IOT devices. While the system is running, these devices will continue to collect data and the expert will provide a second training annotation, i.e. the type of failure that actually occurred, for each data set. Meanwhile, the model predicts the fault type through the first fault recognition processing, calculates the viewpoint label number of each type, and then generates a statistical chart. The second training annotation for each data set may then be compared to the statistical map to derive a confidence coefficient for that data set.

Therefore, by introducing the confidence coefficient of the data set, the confidence level of the model on the prediction result can be better measured, so that the prediction strategy of the model is adjusted, and the accuracy and the efficiency of fault identification are improved.

In some alternative embodiments, the method further comprises, for each of the heating IOT sensing dataset instances, before performing the following steps, respectively: determining representative intelligent heat supply sensing data sets of fault keywords obtained by the first fault identification process from a plurality of intelligent heat supply sensing data set examples of the heat supply IOT sensing data set example cluster according to fault event identification information of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example cluster; for each fault keyword, acquiring an initial characterization vector and a corresponding initial training annotation of the fault keyword, and dynamically optimizing the initial characterization vector by adopting a representative intelligent heat supply sensing data set of the fault keyword to obtain a derivative characterization vector of the fault keyword; for each intelligent heat supply sensing data set example, determining a commonality score between a heat supply fault representation vector of the intelligent heat supply sensing data set example and a derivative representation vector of each fault keyword, and taking an initial training annotation corresponding to a target derivative representation vector with the highest commonality score as an initial data set training annotation of the intelligent heat supply sensing data set example; and aiming at each intelligent heat supply sensing data set example, adopting initial data set training annotation of the intelligent heat supply sensing data set example, and dynamically optimizing second training annotation of the intelligent heat supply sensing data set example to obtain derivative training annotation of the intelligent heat supply sensing data set example.

The fault keywords refer to keywords or phrases for describing the faults of the heating system. For example, "pressure is too high", "temperature is abnormal", etc. Representative intelligent heating sensing data sets refer to data sets that most typically reflect a particular fault keyword among all examples of intelligent heating sensing data sets. Initial token vector and derived token vector: the initial token vector is a preliminary token of the fault keyword, possibly based on some predefined rule or criteria; the derived characterization vector is obtained by learning and optimizing on the basis of the initial characterization vector, and can more accurately represent the characterization vector of the fault keyword. The commonality score is a quantization index for measuring the similarity or matching degree between the heat supply fault characterization vector of the intelligent heat supply sensing data set example and the derivative characterization vector of the fault keyword. The initial data set training annotation is obtained based on the initial training annotation corresponding to the target derived characterization vector with the highest commonality score; the derived training annotation is obtained through dynamic optimization on the basis of the training annotation of the initial data set, and can more accurately reflect the training annotation of fault information in the data set.

For example, IOT sensor data for a heating system is being analyzed. First, a representative intelligent heating sensing data set of each fault keyword is determined according to fault event identification information. Then, for each fault keyword, acquiring an initial characterization vector and an initial training annotation of the fault keyword, and dynamically optimizing the initial characterization vector by using a representative data set of the fault keyword to obtain a derivative characterization vector of the fault keyword. And calculating the commonality scores between the heat supply fault characterization vectors of each intelligent heat supply sensing data set example and the derivative characterization vectors of all fault keywords, and taking the initial training annotation corresponding to the target derivative characterization vector with the highest score as the initial data set training annotation of the data set example. Finally, the second training annotations of the intelligent heating sensing dataset examples are dynamically optimized using the initial dataset training annotations to obtain derivative training annotations.

In this way, fault information in the heating system can be extracted and utilized more accurately. By calculating the commonality score and optimizing the training annotation, the fault condition of the system can be more accurately identified and described, so that the efficiency of fault identification and processing is improved. Meanwhile, the scheme also considers the representativeness of the fault keywords in each intelligent heat supply sensing data set example, so that the fault recognition result has more pertinence and practicability.

In some optional embodiments, the performing, by the fault scenario recognition branch, a second fault recognition process on the heating IOT sensing dataset example in combination with the second training comments of each of the intelligent heating sensing dataset examples in the heating IOT sensing dataset example to obtain a fault recognition perspective example of the heating IOT sensing dataset example includes: combining the derived training notes of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example; the determining a network training error of the heating system fault identification network according to the difference between the fault event identification information and the second training annotation of each intelligent heating sensing data set example and the difference between the fault identification point of view example and the first training annotation of the heating IOT sensing data set example comprises: and determining a network training error of the heating system fault identification network according to the difference between the fault event identification information of each intelligent heating sensing data set example and the derivative training annotation and the difference between the fault identification viewpoint example of the heating IOT sensing data set example and the first training annotation.

For example, there is a heating system from which a collection of heating IOT sensing dataset instances are collected. For each smart heat sensing dataset example, not only are the raw training annotations provided by the expert, but also derivative training annotations are generated by some rule or algorithm. Next, these data are input to the failure scenario recognition branch for the second failure recognition processing, and also processed in combination with the derived training annotations, resulting in failure recognition perspective examples. And finally, determining the network training error of the heating system fault identification network according to the difference between the fault event identification information and the derived training annotation and the difference between the fault identification viewpoint example and the first training annotation.

Therefore, by introducing the derived training annotation, training data can be utilized more comprehensively and deeply, and the accuracy of fault identification is further improved. Meanwhile, the performance of the model can be more accurately estimated and optimized by comparing the difference between the model prediction result and the derived training annotation, so that the running efficiency and stability of the heating system are improved.

In some alternative embodiments, the fault event identification information includes a discrimination likelihood that the smart heating sensor data set instance points to each of the fault keywords; the determining, according to the fault event identification information of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example cluster, a representative intelligent heat supply sensing data set of each fault keyword obtained by the first fault identification process from a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example cluster includes: for each fault keyword, the following steps are respectively implemented: determining target intelligent heat supply sensing data set examples which point to the fault keywords and have the possibility of judging from high to low target numbers from a plurality of intelligent heat supply sensing data set examples of the heat supply IOT sensing data set example cluster; and taking the target intelligent heat supply sensing data set of the target number as a representative intelligent heat supply sensing data set of the target number of the fault keywords.

Wherein the discrimination probability is a probability value representing the probability that the model predicts that the example of the intelligent heat supply sensing data set points to a certain fault keyword. The larger the value, the more likely the data set instance corresponds to the fault keyword. The target number refers to the number of data set examples with the highest discrimination possibility selected from all intelligent heat supply sensing data set examples.

For example, IOT sensor data for a heating system is being analyzed. First, the discrimination possibility of each intelligent heat supply sensing data set example pointing to each fault keyword is obtained through the first fault identification processing. Then, for each fault keyword, a plurality of (i.e. target number of) examples of intelligent heat supply sensing data sets with highest discrimination possibility are selected and used as representative intelligent heat supply sensing data sets of the fault keyword.

The above-described embodiments provide a method of selecting representative intelligent heating sensing data sets based on discrimination possibilities, which facilitates better understanding and identification of various fault keywords. Meanwhile, the scheme can more accurately locate and describe the faults of the heating system, so that the efficiency and the accuracy of fault processing are improved. In addition, the complexity and the calculated amount of the model can be adjusted according to actual needs by dynamically selecting the target number, so that the model can ensure good performance and cannot excessively consume calculation resources.

In some optional embodiments, the number of the representative intelligent heat supply sensing data sets is X, where X is an integer greater than 1, and dynamically optimizing the initial characterization vector to obtain a derivative characterization vector of the fault keyword by using the representative intelligent heat supply sensing data set of the fault keyword includes: dynamically optimizing the initial characterization vector by adopting the 1 st representative intelligent heat supply sensing data set of the fault keyword to obtain a1 st initial characterization vector to be processed of the fault keyword; dynamically optimizing the (u-1) th to-be-processed initial characterization vector of the fault keyword by adopting the (u) th representative intelligent heat supply sensing data set of the fault keyword to obtain the (u) th to-be-processed initial characterization vector of the fault keyword, wherein u is greater than 0 and not greater than X; and performing self-adding 1 circulation on the u to obtain an X-th to-be-processed initial characterization vector of the fault keyword, and taking the X-th to-be-processed initial characterization vector of the fault keyword as a derivative characterization vector of the fault keyword.

The initial characterization vector to be processed is a vector generated in the middle step of the optimization process, and is a result obtained by performing one or more dynamic optimization on the initial characterization vector. This vector is used in the next optimization process until all representative intelligent heating sensing data sets have been used.

For example, there is one fault keyword and 3 (i.e., x=3) representative intelligent heating sensing data sets. First, the 1 st representative data set is used to dynamically optimize the initial characterization vector, and the 1 st initial characterization vector to be processed is obtained. The 1 st initial characterization vector to be processed is then dynamically optimized using the 2 nd representative data set to obtain the 2 nd initial characterization vector to be processed. Next, the 3 rd representative data set is used to dynamically optimize the 2 nd initial characterization vector to be processed, resulting in the 3 rd initial characterization vector to be processed. And finally, taking the 3 rd initial characterization vector to be processed as a derivative characterization vector of the fault key word.

According to the technical scheme, dynamic optimization is performed step by step, the information of each representative intelligent heat supply sensing data set is fully utilized, and the derivative characterization vector of the fault keyword can be obtained more accurately. This stepwise optimization approach also helps to avoid overfitting, resulting in better generalization of the model over unknown data. In general, the technical scheme can improve the accuracy of fault identification and the stability of the model.

In some optional embodiments, the number of the representative intelligent heat supply sensing data sets is a plurality, the adopting the representative intelligent heat supply sensing data sets of the fault keywords dynamically optimizes the initial characterization vector to obtain derivative characterization vectors of the fault keywords, and includes: obtaining representative heat supply fault representation vectors of each representative intelligent heat supply sensing data set of the fault keywords, and determining heat supply fault average representation vectors of a plurality of representative heat supply fault representation vectors; acquiring a first characterization vector confidence coefficient of the heat supply fault average characterization vector and a second characterization vector confidence coefficient of the initial characterization vector; and performing error arrangement on the heat supply fault average characterization vector and the initial characterization vector according to the first characterization vector confidence and the second characterization vector confidence to obtain the derivative characterization vector of the fault keyword.

The first token vector confidence and the second token vector confidence are indexes for evaluating the accuracy of the token vector. The first token vector confidence corresponds to the accuracy of the heating fault average token vector, and the second token vector confidence corresponds to the accuracy of the initial token vector.

For example, there is a heating system where a large amount of data is collected by sensors. A plurality of representative intelligent heat supply sensing data sets are selected, and an initial characterization vector is calculated for each fault keyword. Then, a heat supply failure characterization vector for each representative data set is obtained, and their average value is calculated, resulting in a heat supply failure average characterization vector. The first token vector confidence of this average token vector and the second token vector confidence of the initial token vector are calculated simultaneously. And finally, according to the two confidence coefficients, carrying out error arrangement on the average characterization vector and the initial characterization vector of the heat supply faults, and obtaining the derivative characterization vector of each fault keyword.

According to the technical scheme, the initial characterization vector is dynamically optimized, so that the characteristics of data can be better captured, and the accuracy of fault identification is improved. Meanwhile, the stability and reliability of the optimization process can be ensured by calculating and considering the confidence coefficient of the characterization vector. In addition, the representative data set is dynamically selected, so that the method has good flexibility and can adapt to various different scenes and requirements.

In some alternative embodiments, the dynamically optimizing the second training annotation of the smart heating sensing dataset with the initial dataset training annotation of the smart heating sensing dataset, resulting in a derivative training annotation of the smart heating sensing dataset, comprises: acquiring a first training annotation confidence coefficient of the training annotation of the initial data set and a second training annotation confidence coefficient of the second training annotation; and performing error arrangement on the training notes of the initial data set and the second training notes according to the first training note confidence level and the second training note confidence level to obtain derivative training notes of the intelligent heat supply sensing data set example.

Wherein the first training annotation confidence measure indicates the confidence level of the model for the first training annotation (i.e., the initial data set training annotation), and a larger value indicates that the model is more confident in the prediction of this training annotation. The second training annotation confidence represents the confidence of the model for the second training annotation. Error management is a process that adjusts or corrects two training annotations based on their confidence levels to obtain more accurate derived training annotations.

For example, there is an example of a smart heating sensing dataset having an initial dataset training annotation and a second training annotation. The confidence coefficient of the two training notes is firstly obtained, then error arrangement is carried out according to the two confidence coefficients, and finally the derived training notes of the data set example are obtained.

It can be seen that by using the first training annotation confidence level and the second training annotation confidence level, the training annotations can be more accurately evaluated and adjusted, resulting in more accurate derived training annotations. The dynamic optimization mode can fully utilize the prediction information of the model, improve the quality of training notes, and further improve the accuracy of fault identification and the stability of the model.

On the above, there is provided a data processing system comprising:

The description of the above functional modules may refer to the description of the method shown in fig. 1, and will not be repeated herein.

On the basis of the above, a data processing system is provided comprising a processor and a memory in communication with each other, said processor being arranged to retrieve a computer program from said memory and to implement the above-mentioned method by running said computer program.

On the basis of the above, a computer readable storage medium is provided, on which a computer program is stored, which computer program, when run, implements the method described above.

The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and variations of the present application will be apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the application are to be included in the scope of the claims of the present application.

Claims

1. A method for processing data of an intelligent heating system, applied to a data processing system, the method comprising:

performing second fault identification processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a global fault identification view of the target heat supply IOT sensing data set;

the disassembly of the target heat supply IOT sensing dataset to obtain a plurality of intelligent heat supply sensing datasets comprised by the target heat supply IOT sensing dataset comprises:

Utilizing the mask interval to activate the data text mask box, and disassembling the target heat supply IOT sensing data set to obtain a plurality of intelligent heat supply sensing data sets included in the target heat supply IOT sensing data set;

The fault representation vector mining is performed on each intelligent heat supply sensing data set to obtain a heat supply fault representation vector of each intelligent heat supply sensing data set, and the heat supply fault representation vector mining comprises the following steps:

Performing fault characterization vector mining of a second feature scale on the to-be-processed heat supply fault characterization vector spectrum to obtain a heat supply fault characterization vector spectrum of the target heat supply IOT sensing dataset, wherein the heat supply fault characterization vector spectrum comprises the heat supply fault characterization vectors of the second feature scale of each intelligent heat supply sensing dataset, and the second feature scale is smaller than the first feature scale;

the determining the confidence coefficient of the intelligent heat supply sensing data set according to the local fault identification viewpoint of the intelligent heat supply sensing data set comprises the following steps:

Taking the difference as a confidence coefficient of the intelligent heat supply sensing data set;

Wherein, according to the heat supply fault characterization vector and the confidence coefficient of the intelligent heat supply sensing data set, determining the heat supply fault attention vector of the intelligent heat supply sensing data set includes: weighting the heat supply fault characterization vector of the intelligent heat supply sensing data set and the confidence coefficient of the intelligent heat supply sensing data set to obtain the heat supply fault attention vector of the intelligent heat supply sensing data set;

The first fault identification processing is realized through a fault event identification branch of a heat supply system fault identification network, and the second fault identification processing is realized through a fault scene identification branch of the heat supply system fault identification network.

2. The method of claim 1, wherein performing a second fault identification process on the target heat supply IOT sensing dataset according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing datasets to obtain a global fault identification view of the target heat supply IOT sensing dataset, comprises:

3. The method according to claim 1, wherein the method further comprises:

4. A method according to claim 3, wherein said determining a network training error of said heating system fault identification network based on said fault event identification information and said second training annotation differences for each of said intelligent heating sensing dataset examples and said fault identification perspective examples and said first training annotations differences for said heating IOT sensing dataset examples comprises:

5. A method according to claim 3, wherein said performing, by said fault event identification branch, a first fault identification process on each of said intelligent heat supply sensing data set instances in each of said heat supply IOT sensing data set instances, respectively, to obtain fault event identification information for each of said intelligent heat supply sensing data set instances, comprises:

Performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch according to the heat supply fault strengthening characterization vector examples of a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example;

wherein said determining a data set confidence coefficient for each of said intelligent heat supply sensing data set examples from said second training annotations for each of said intelligent heat supply sensing data set examples in said heat supply IOT sensing data set examples comprises: for each intelligent heat supply sensing data set example in the heat supply IOT sensing data set examples, respectively implementing the following steps: acquiring the viewpoint label number of the first fault identification process, and determining a statistical graph of the viewpoint label number; determining a distinction between the second training annotation of the smart heating sensor dataset example and the statistical map; the differences are taken as data set confidence coefficients for the smart heating sensor data set example.

6. A method according to claim 3, wherein the method further comprises, for each of the heating IOT sensing dataset instances, before performing the steps of:

For each intelligent heat supply sensing data set example, adopting initial data set training annotation of the intelligent heat supply sensing data set example, and dynamically optimizing second training annotation of the intelligent heat supply sensing data set example to obtain derivative training annotation of the intelligent heat supply sensing data set example;

the second training annotation combining each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example performs second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example, and the method comprises the following steps: combining the derived training notes of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example, and performing second fault identification processing on the heat supply IOT sensing data set example through the fault scene identification branch to obtain a fault identification viewpoint example of the heat supply IOT sensing data set example; the determining a network training error of the heating system fault identification network according to the difference between the fault event identification information and the second training annotation of each intelligent heating sensing data set example and the difference between the fault identification point of view example and the first training annotation of the heating IOT sensing data set example comprises: determining a network training error of the heating system fault identification network according to the difference between the fault event identification information and the derived training annotations of each intelligent heating sensing data set example and the difference between the fault identification point of view example and the first training annotations of the heating IOT sensing data set example;

Wherein the fault event identification information comprises a discrimination possibility that the intelligent heat supply sensing data set example points to each fault keyword; the determining, according to the fault event identification information of each intelligent heat supply sensing data set example in the heat supply IOT sensing data set example cluster, a representative intelligent heat supply sensing data set of each fault keyword obtained by the first fault identification process from a plurality of intelligent heat supply sensing data set examples in the heat supply IOT sensing data set example cluster includes: for each fault keyword, the following steps are respectively implemented: determining target intelligent heat supply sensing data set examples which point to the fault keywords and have the possibility of judging from high to low target numbers from a plurality of intelligent heat supply sensing data set examples of the heat supply IOT sensing data set example cluster; taking the target intelligent heat supply sensing data set example of the target number as a representative intelligent heat supply sensing data set of the target number of the fault keywords;

The number of the representative intelligent heat supply sensing data sets is X, X is an integer larger than 1, the representative intelligent heat supply sensing data sets adopting the fault keywords dynamically optimize the initial characterization vector to obtain derivative characterization vectors of the fault keywords, and the method comprises the following steps: dynamically optimizing the initial characterization vector by adopting the 1 st representative intelligent heat supply sensing data set of the fault keyword to obtain a1 st initial characterization vector to be processed of the fault keyword; dynamically optimizing the (u-1) th to-be-processed initial characterization vector of the fault keyword by adopting the (u) th representative intelligent heat supply sensing data set of the fault keyword to obtain the (u) th to-be-processed initial characterization vector of the fault keyword, wherein u is greater than 0 and not greater than X; the automatic adding 1 loop is carried out for the u, an X-th to-be-processed initial characterization vector of the fault keyword is obtained, and the X-th to-be-processed initial characterization vector of the fault keyword is used as a derivative characterization vector of the fault keyword; or the number of the representative intelligent heat supply sensing data sets is a plurality of, the representative intelligent heat supply sensing data sets adopting the fault keywords dynamically optimizes the initial characterization vector to obtain derivative characterization vectors of the fault keywords, and the method comprises the following steps: obtaining representative heat supply fault representation vectors of each representative intelligent heat supply sensing data set of the fault keywords, and determining heat supply fault average representation vectors of a plurality of representative heat supply fault representation vectors; acquiring a first characterization vector confidence coefficient of the heat supply fault average characterization vector and a second characterization vector confidence coefficient of the initial characterization vector; performing error arrangement on the heat supply fault average characterization vector and the initial characterization vector according to the first characterization vector confidence and the second characterization vector confidence to obtain a derivative characterization vector of the fault keyword;

wherein, the training annotation of the initial data set using the smart heat supply sensing data set example dynamically optimizes the second training annotation of the smart heat supply sensing data set example to obtain the derived training annotation of the smart heat supply sensing data set example, comprising: acquiring a first training annotation confidence coefficient of the training annotation of the initial data set and a second training annotation confidence coefficient of the second training annotation; and performing error arrangement on the training notes of the initial data set and the second training notes according to the first training note confidence level and the second training note confidence level to obtain derivative training notes of the intelligent heat supply sensing data set example.

7. A data processing apparatus, comprising:

The second recognition module is used for carrying out second fault recognition processing on the target heat supply IOT sensing data set according to the heat supply fault attention vectors of the plurality of intelligent heat supply sensing data sets to obtain a global fault recognition view of the target heat supply IOT sensing data set;

8. A data processing system comprising a processor and a memory in communication with each other, the processor being arranged to retrieve a computer program from the memory and to implement the method of any of claims 1-6 by running the computer program.