CN120541585B

CN120541585B - Safety monitoring method and monitoring system for oil well construction

Info

Publication number: CN120541585B
Application number: CN202511036660.XA
Authority: CN
Inventors: 郑国林; 方佳伟; 李留; 刘一鸣; 彭书友; 周燕武; 罗成; 王翱宇
Original assignee: Hangzhou Harmonycloud Technology Co Ltd
Current assignee: Hangzhou Harmonycloud Technology Co Ltd
Priority date: 2025-07-28
Filing date: 2025-07-28
Publication date: 2025-10-21
Anticipated expiration: 2045-07-28
Also published as: CN120541585A

Abstract

The invention discloses a safety monitoring method and a safety monitoring system for oil well construction, which belong to the technical field of electric digital data processing, wherein the method comprises the following steps of collecting monitoring data of the oil well construction; extracting features from the monitoring data, fusing the features and the monitoring data to obtain fused features, predicting detection values of the fused features through a prediction model based on machine learning to obtain predicted values, calculating first abnormal degrees according to the predicted values and the detected values, obtaining second abnormal degrees and safety scores according to the predicted values and the detected values if the first abnormal degrees exceed a first threshold value, and carrying out abnormal alarming according to the safety scores and the second threshold value. The multi-mode feature fusion is suitable for complex and changeable construction environments, large-scale and high-dimensional monitoring data, the calculation efficiency is improved, and the accuracy and the reliability of monitoring are improved by carrying out anomaly detection and anomaly alarm based on the predicted value and the detected value of the prediction model.

Description

Safety monitoring method and monitoring system for oil well construction

Technical Field

The invention relates to the technical field of electric digital data processing, in particular to a safety monitoring method and a safety monitoring system for oil well construction.

Background

Sensor networks and industrial internet of things systems at construction sites can generate large amounts of real-time data, such as pressure, temperature, flow, vibration, equipment running states, environmental conditions and other multidimensional data, which are the cores for ensuring the safety and efficiency of oil well construction. These data are of great importance to construction safety, efficiency and quality. However, the construction site environment is complex, and the sensor may be affected by interference or faults in long-term operation, so that data are abnormal or lost, and potential risks are brought to construction safety.

For example, extreme environmental conditions (such as high temperature, high pressure, or corrosive media) may affect the accuracy or lifetime of the sensor, thereby causing data inconsistencies. In addition, there are various dynamic factors in the construction process, such as pressure fluctuation, equipment vibration or working condition change, and these factors may introduce noise data, further increasing the difficulty of data monitoring.

With the development of large model technology, large models based on deep learning and reinforcement learning are capable of generating test data simulating the oil well construction process, which are generally used to simulate extreme scenes or to test the robustness and emergency handling capability of the construction system. However, large model generated data may have potential anomalies or logical contradictions due to model bias or input data imperfections in the generation process, which further increases the requirements for data security monitoring.

Therefore, achieving efficient anomaly detection and security assessment in complex, diverse environments is an important research direction.

Disclosure of Invention

Aiming at the technical problems in the prior art, the invention provides the safety monitoring method and the safety monitoring system for the oil well construction, which improve the efficiency and the safety of monitoring the oil well construction data.

The invention discloses a safety monitoring method for oil well construction, which comprises the following steps of collecting monitoring data of the oil well construction, extracting features from the monitoring data, fusing the features and the monitoring data to obtain fused features, predicting detection values of the fused features through a prediction model based on machine learning to obtain predicted values, calculating first abnormal degrees according to the predicted values and the detected values, judging whether the first abnormal degrees exceed a first threshold value, obtaining second abnormal degrees and safety scores according to the predicted values and the detected values if the first abnormal degrees exceed the first threshold value, and carrying out abnormal alarm according to the safety scores and the second threshold value.

Preferably, the extracted features include any one or combination of temperature, pressure, flow, humidity, rate of pressure change, temperature gradient, and flow fluctuation amplitude;

the fusion features include any one or combination of data source features, feature level features and decision level features;

The monitoring data is also preprocessed, wherein the preprocessing comprises data cleaning, noise removal, outlier identification and data normalization.

Preferably, the rate of pressure change is expressed as:

;

wherein P (t) is the pressure value at time t, In order to provide for the time interval of time,As the value of the pressure change, a pressure change value,Is the rate of change of pressure;

temperature gradient Expressed as:

;

Wherein, T (T2) and T (T1) are temperature values of time T1 and T2 respectively, and d is the distance between the temperature sensors;

Flow fluctuation amplitude Expressed as:

;

Wherein, Q _max and Q _min are the maximum flow value and the minimum flow value in the time window, respectively;

the data source signature F _fusion is represented as:

;

Where w _i is the weight of the data sources, d _ik represents the kth data point in the ith data source, m is the total number of data points, and n1 is the total number of data sources;

feature level feature X _fusion is represented as:

;

Wherein X _fusion is a feature level feature, X _j is a feature vector of a j-th data source, w _j is a weight of the feature vector, and n2 is a total number of feature vectors;

the decision level feature Y _fusion is represented as:

;

Where Y _q is the decision output of the q-th data source, w _q is the weight of the decision output, 。

Preferably, the weights are updated based on a bayesian formulation expressed as:

;

Wherein w is represented as a weight, w is selected from the group consisting of the weight of the data source, the weight of the feature vector and the weight of the decision output, For the posterior probability of the weights in the case of data D,For the likelihood of data D under weight, P (w) is the a priori distribution of weights and P (D) is the total probability of data D.

Preferably, the method of machine learning comprises a support vector machine, a decision tree or a neural network,

The neural network includes a transducer.

Preferably, the first anomaly degree is calculated by:

;

wherein A (t) is the first anomaly of time step t, X _i2 (t) represents the input data/detection value of the i2 nd sensor at time step t, D2 is the total number of sensors for predicting the predicted value of the model at time step t;

The first threshold P1 is defined as:

;

Wherein A _max represents the maximum outlier in the history data, Is a safety factor.

Preferably, the second degree of abnormality is expressed as:

;

Wherein A (X _i3) is the detection value of the ith 3 time windows of X _i3, In order to be able to predict the value,Standard deviation of historical data;

The security evaluation S is expressed as:

;

Wherein, the The weight coefficient, denoted as the impact of the second anomaly on the security score, l is denoted as the number of data bars under the current time window i 3.

Preferably, the level determination method of the abnormality alarm is as follows:

;

where L is denoted as the level of the alarm.

Preferably, the boundary of the threshold value or the threshold value interval is updated, and the updated threshold value or threshold value interval is expressed as:

;

wherein T is an initial threshold or a threshold interval boundary, a correlation coefficient R represents the matching degree of an alarm event and a real abnormality in historical data, and a historical false alarm rate Wb represents the past alarm error proportion;

The correlation coefficient R expression is as follows:

;

The alarm control method comprises the steps of triggering an alarm, wherein the probability of actually generating the alarm after the alarm is triggered is represented, the reliability of the alarm is represented, the probability of generating the alarm when the alarm is generated, namely the proportion of historically effective alarms is represented, P (E) is the prior probability of the abnormality, namely the overall occurrence frequency of the abnormality, and P (A) is the overall occurrence frequency of the alarm.

The invention also provides a monitoring system for realizing the safety monitoring method, which comprises an acquisition module, a feature extraction module, a fusion module, an abnormality detection module and an alarm module;

the system comprises an acquisition module, a feature extraction module, a fusion module and a feature analysis module, wherein the acquisition module is used for acquiring monitoring data of oil well construction;

The anomaly detection module is used for analyzing the fusion characteristics based on the prediction model to obtain a predicted value, and obtaining a first anomaly degree according to the predicted value and the detection value;

The alarm module is used for obtaining second abnormality degree and safety score according to the predicted value and the detected value, and carrying out abnormal alarm according to the safety score and the second threshold value.

Compared with the prior art, the method has the beneficial effects that complex and changeable construction environments are adapted through multi-mode feature fusion, large-scale and high-dimensional monitoring data are adapted, the calculation efficiency is improved, and the accuracy and the reliability of monitoring are improved through abnormality detection and abnormality alarm based on the predicted value and the detected value of the prediction model.

Drawings

FIG. 1 is a flow chart of a safety monitoring method of the oil well construction of the present invention;

fig. 2 is a logic block diagram of a monitoring system.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

The invention is described in further detail below with reference to the attached drawing figures:

The first aspect of the invention provides a safety monitoring method for oil well construction, comprising the following steps:

And S1, collecting monitoring data of oil well construction and preprocessing.

In the oil well construction site, data acquisition is achieved by deploying a variety of sensors and monitoring equipment. The sensors comprise a pressure sensor, a temperature sensor, a flowmeter, a vibration sensor and an environment monitoring instrument, and are respectively used for monitoring key parameters of the inside and the surrounding environment of the oil well in real time.

And S2, extracting features from the monitoring data, and fusing the features and the monitoring data to obtain fused features.

And step S3, constructing the training set according to the fusion characteristics.

And step S4, training the training set based on a machine learning method to obtain a prediction model.

And S5, obtaining a predicted value through a predicted model.

And S6, obtaining a first degree of anomaly according to the predicted value and the detected value.

And S7, judging whether the first abnormality degree exceeds a first threshold value.

If not, monitoring the monitoring data continuously.

If yes, step S8 is executed, and the second anomaly degree and the security score are obtained according to the predicted value and the detected value.

And step S9, carrying out abnormal alarm according to the safety score and the second threshold value.

The multi-mode feature fusion is suitable for complex and changeable construction environments, large-scale and high-dimensional monitoring data, the calculation efficiency is improved, and the accuracy and the reliability of monitoring are improved by carrying out anomaly detection and anomaly alarm based on the predicted value and the detected value of the prediction model.

Examples. The method realizes the real-time monitoring and abnormal alarm of the oil well construction data, and one of the core steps is the data acquisition and preprocessing in the step S1. The preprocessing comprises data cleaning, noise removal, outlier identification and data normalization.

This step is the basis of the whole monitoring system, and aims to ensure that the data acquired from multiple sources has high quality, integrity and consistency, so as to provide reliable input for subsequent dynamic anomaly detection, multi-mode data fusion and alarm mechanisms. The data acquisition and preprocessing module is not only responsible for acquiring and cleaning data, but also normalizes the data and extracts the characteristics through a series of algorithms, so that the monitoring system can accurately and effectively identify potential safety hazards in a complex construction environment. In the oil well construction site, data acquisition is achieved by deploying a variety of sensors and monitoring equipment. The sensors comprise a pressure sensor, a temperature sensor, a flowmeter, a vibration sensor, an environment monitoring instrument and the like, and are respectively used for monitoring key parameters of the inside and the surrounding environment of the oil well in real time. The sensors can transmit the acquired data in real time through an industrial Internet of things network, and the acquired data are uploaded to a central data processing system and the like. Meanwhile, test data generated by the large model are also input into the system through a preset interface and are used for simulating extreme working conditions and abnormal scenes so as to supplement and enrich the monitoring range of real-time data.

The raw data collected often contains noise and outliers that may be introduced due to sensor failures, environmental disturbances, or data transmission errors. In practical application, real-time data of a construction site and test data generated by a large model are required to be combined to cover various scenes in the construction process. However, how to effectively fuse these two types of data and perform unified security monitoring is one of the current technical difficulties due to different sources and different characteristics of the two types of data. Traditional rule-driven data monitoring methods rely on static safety rules or empirical knowledge and cannot be dynamically adapted to complex and changeable construction environments. The method has weak detection capability on burst abnormal or unknown modes, and is easy to cause a monitoring blind area. In addition, when the traditional method is used for processing large-scale and high-dimensional data, the problem of low calculation efficiency is often faced, and the requirement of real-time monitoring is difficult to meet.

Therefore, the data is first subjected to cleaning and denoising. To effectively remove high frequency noise, a gaussian filter may be used for smoothing. The Gaussian filter effectively suppresses the influence of high-frequency noise by performing weighted average on each data point and the data of the neighborhood around the data point, and the weight is determined by a Gaussian function. The specific formula is as follows:

(1);

Wherein X (t) represents the data value of time t, K is the length of the time window i, which is the standard deviation of the filter. Through the gaussian filtering process, the system is able to preserve the dominant trend of the data while reducing fluctuations caused by environmental noise or device errors.

After the denoising process, the abnormal value in the data is further identified and removed. Because extreme values caused by sensor faults or instantaneous interference can exist in the construction data, the system adopts a statistical method based on a box diagram to detect the abnormal values. First, the system calculates a first quartile Q ₁ and a third quartile Q ₃ of the dataset, and a quartile spacing (iqr=q ₃- Q₁). Defining an outlier range as:

Outlier range= [ Q ₁ 1.5·IQR, Q₃ + 1.5·IQR] (2);

Data points outside this range are considered outliers and are culled. Abnormal values that deviate significantly from the normal data distribution can be effectively identified and removed while preserving the overall characteristics and distribution morphology of the data set. The data of different sources and different types can be compared and analyzed in a unified feature space, and the data can be normalized. And specifically, a minimum-maximum normalization method is adopted to compress all data to be within the range of 0 and 1, so that the influence caused by the difference of the data magnitude of different sensors is eliminated. The normalization formula is as follows:

(3);

wherein X _min and X _max are the minimum and maximum values, respectively, in the dataset. Through normalization processing, the system ensures that all input data X are subjected to subsequent fusion and analysis under the same scale, and improves the efficiency and accuracy of algorithm processing.

The purpose of feature extraction in step S2 is to extract key indexes reflecting construction states and potential risks from the original data, and construct a high-dimensional feature vector for use in a subsequent anomaly detection algorithm.

Specifically, in addition to basic characteristics such as temperature, pressure, flow, humidity and the like, key characteristics such as pressure change rate, temperature gradient and flow fluctuation range are extracted.

The pressure change rate is an important index for reflecting the dynamic change of the pressure in the oil well, and the calculation formula is as follows:

(4);

wherein P (t) is the pressure value at time t, In order to provide for the time interval of time,As the value of the pressure change, a pressure change value,Is the rate of change of pressure. This feature reveals the instantaneous trend of pressure, and is of great importance for detecting pressure anomalies.

Temperature gradientFor reflecting the rate of change of the temperature of the well and its equipment, the calculation formula is as follows:

(5);

wherein, T (T2) and T (T1) are temperature values of time T1 and T2, respectively, and d is the distance between the temperature sensors. Changes in the temperature gradient may indicate overheating or uneven cooling of the device, indicating potential device failure or environmental problems.

Flow fluctuation amplitudeThe calculation formula of the important features reflecting the fluid dynamics of the oil well is as follows:

(6);

Wherein Q _max and Q _min are the maximum flow value and the minimum flow value, respectively, within the time window. An increase in the magnitude of the flow fluctuation may be indicative of fluid flow instability or pipe blockage.

Through the feature extraction, a plurality of key parameters are converted into high-dimensional feature vectors, and a rich information basis is provided for subsequent dynamic anomaly detection and multi-mode data fusion. The process not only improves the interpretability of the data, but also enhances the identification capability of the anomaly detection algorithm on potential risks in a complex construction environment. Data acquisition and preprocessing play a vital role in the present invention. The high quality and consistency of the input data can be ensured through efficient denoising, accurate outlier rejection, a unified normalization method and extraction of key features. A solid foundation is laid for successful implementation of the intelligent data safety monitoring method, so that potential safety risks in the oil well construction process can be accurately identified and early-warned under the reliable data support by subsequent dynamic anomaly detection and multi-mode feature fusion.

The feature fusion of the step S2 is a key link for realizing high-efficiency data analysis and anomaly detection. The multi-mode data fusion method has the main effects that multi-mode data from different data sources are fused, and monitoring accuracy and reliability are improved. In the oil well construction process, the data sources are various, and the data (such as sensor data of pressure, temperature, flow, vibration and the like) collected in real time on site, simulation data and historical data generated by a large model and the like are covered. Because these data types have different sources, scales, and formats, an efficient fusion mechanism is needed to fully understand the real-time status of well construction from multiple dimensions. The design of the multi-mode data fusion module aims at mining potential relations among different data types through deep fusion of data, so that the detection capability of complex abnormal conditions is improved. By processing various heterogeneous data sources, data redundancy is eliminated, and complementarity of data is enhanced, so that a subsequent anomaly detection and early warning system is helped to identify potential risks more accurately.

In specific feature fusion, a fusion strategy based on weighted summation is adopted, and the data source features F _fusion, the feature level features X _fusion and the decision level features Y _fusion. are fused, so that different weights can be distributed for each data type according to the credibility and the importance of different data sources, and further, more accurate comprehensive evaluation is realized.

Given a set of multimodal data, each item of data D _i corresponds to a weight W _i that reflects the importance of that data in the final fusion. The goal of data fusion is to calculate a weighted average or weighted sum such that each data source's contribution to the final result is proportional to its weight. For each data source = {,.., }, the fusion result is expressed as:

(7);

Where w _i is the weight of the data source, d _ik represents the kth data point in the ith data source, and m is the total number of data points. The final multimodal data fusion result F _fusion is the sum of all data source weighted sums:

(8);

In this process, w _i is a weight set according to the importance and reliability of the data sources, which can be generally determined by historical data analysis or expert evaluation, and n1 is represented as the total number of data sources. The concept of weighted summation ensures different influence of each data source in the fusion process, so that a fusion result comprehensively considering a plurality of factors is obtained.

The determination of the weights w _i is typically a dynamic process. In order to be able to more accurately reflect the impact of each data source, the weights may be dynamically adjusted in combination with the historical performance and reliability of the data. A common way is to use the accuracy, stability, and correlation of historical data of the data source to adjust the weights. For example, when a data source (such as pressure sensor data) exhibits a strong abnormality recognition capability, its weight may be appropriately increased, and when a data source has an error or a large deviation, its weight may be appropriately decreased. In practical applications, the weights may be optimized by a learning algorithm, such as using bayesian inference methods to infer the trustworthiness of each data source from historical data, and dynamically adjusting the weights using the trustworthiness. Specifically, the weights are updated using a bayesian formula by calculating the contribution of the data source in the historical monitoring and the correlation with other data sources:

(9);

Wherein, the For the posterior probability of weight w _i in the case of data D,For the likelihood of data D under weight w _i, P (w _i) is the prior distribution of weights and P (D) is the total probability of data D.

Multimodal data fusion is not only a direct weighted average of the raw data, but can also be fused at the feature level and decision level. The feature level fusion is to perform joint modeling on features (such as pressure change rate, temperature gradient, flow fluctuation and the like) of different data sources, and more comprehensive features are extracted through feature fusion. This process can be modeled by multiple linear regression or neural networks, etc. The features of each data source may be represented as a vector X _j=[x_j1,x_j2,...,x_jn2, then the feature level fusion process may be performed by way of weighted features:

(10);

Where X _fusion is the feature level feature, X _j is the feature vector of the j-th data source, w _j is the weight of the feature vector, and n2 is the total number of feature vectors. The fused feature vector X _fusion will contain information from multiple data sources and be processed by a subsequent anomaly detection module.

The decision level fusion is performed on the basis of the detection result of each data source. When each data source determines whether an anomaly exists through an independent detection model, the results can be combined through voting or weighted averaging. Assuming that the decision output of each data source is where 0 represents no anomalies and 1 represents anomalies, the decision-level fusion process can be expressed as:

(11);

According to the integrated decision value Y _fusion,Y_q being the decision output of the q-th data source, w _q is the weight of the decision output. The system will determine if an abnormal situation exists. If it is Above a certain threshold (typically 0.5), anomalies are considered to be present, otherwise data is considered normal. w _k and w _q are similarly updated by a bayesian formula, which can be expressed as:

(12);

Where w is denoted as a weight selected from the group consisting of the weight of the data source, the weight of the feature vector and the weight of the decision output, For the posterior probability of the weights in the case of data D,For the likelihood of data D under weight, P (w) is the a priori distribution of weights and P (D) is the total probability of data D.

And obtaining the comprehensive representation of the multi-mode data through the fusion of the feature level and the decision level. Based on the comprehensive representation, comprehensive evaluation and anomaly detection are carried out on the safety state of oil well construction. The multi-layer perceptron model in deep learning can be used for mapping field data and test data to a unified feature space to generate a fusion feature representation. The fused data can capture the internal association between the two types of data, and the accuracy and the applicability of anomaly detection are improved.

The fused data can be trained and classified by using a machine learning model (such as a support vector machine, a decision tree or a neural network), so that whether the current oil well construction state is abnormal or not can be automatically judged. The result of the anomaly detection provides input for a subsequent alarm mechanism, so that an alarm can be timely sent out in the early stage of anomaly occurrence, and potential safety accidents are avoided. The design and implementation of the multi-mode data fusion module enable the invention to effectively integrate data from different sources and perform the fusion operation of weighted summation or weighted average according to the weight, the historical performance and the credibility of each data source. Through the fusion of the feature level and the decision level, the system can evaluate the safety state of oil well construction more comprehensively and accurately, and provides accurate data support for subsequent abnormality detection and alarm mechanisms.

With the development of artificial intelligence technology, intelligent algorithms based on machine learning and deep learning have shown great potential in the field of data security monitoring. For example, the anomaly detection algorithm can automatically identify potential anomaly patterns by learning the distribution characteristics of historical data, and the timing analysis technique can capture the dynamic change rules of the data for predicting the potential anomaly trend. However, the application of these techniques in the field of oil well construction is still in the preliminary stage. How to design an efficient and accurate algorithm aiming at specific characteristics of construction site and large model generated data and simultaneously ensure reliability and interpretability of monitoring results is a technical challenge to be solved urgently.

In step S5, the prediction model is another core part of the present invention, and is responsible for real-time monitoring and anomaly identification of the oil well construction data. During the well construction process, a potential anomaly pattern is automatically identified and an alarm mechanism is triggered. With the rapid increase of data volume in the oil well construction process, the traditional static monitoring method cannot effectively identify complex and changeable abnormal modes.

The invention particularly provides a prediction model based on a transducer network, so as to process high-dimensional and multi-mode time series data more accurately and efficiently.

The oil well construction data has the characteristics of various data types, frequent generation, large fluctuation and unusual change of abnormal modes. In order to discover potential security hazards in time, the data stream must be monitored dynamically and continuously. Traditional anomaly detection methods, such as a control chart method based on statistical analysis, a rule-based detection method and the like, have strong limitations, and are difficult to cope with the complexity and diversity of oil well construction data. Especially, when the data amount increases, the conventional method is susceptible to data noise and change patterns, resulting in false alarm and false alarm. And a prediction model based on a transducer is introduced, dynamic data of oil well construction is accurately modeled through the strong time sequence modeling capability and a self-attention mechanism of the transducer, and abnormal changes in a data stream are detected in real time. The transducer can automatically capture long-term dependency relationship in data, so that the defect of the traditional method in processing long-term dependency data is avoided, and the accuracy of anomaly detection is improved.

The self-attention mechanism of Transforme plays a key role in the dynamic anomaly detection process. Compared with the traditional Recurrent Neural Network (RNN), the transducer is independent of a serialized calculation process, and can process input data in parallel, so that the processing efficiency is remarkably improved, and particularly when large-scale oil well construction data are processed.

First, data X (t) generated during the well construction process is a time series data set, where t represents time steps and X (t) represents sensor input data (e.g., temperature, pressure, flow, etc.) at time point t. The data stream is typically multi-dimensional, with each dimension potentially representing different sensor information, and therefore such multi-modal data needs to be fused and processed. In the input and embedding layer of the transducer, the original time series data is processed and converted, so that the input data is ensured to meet the requirements of a model. Assuming T time steps, the input data for each time step is a d-dimensional vector representing the values of all sensors at that time. The input of the model may be represented as a matrix. In order for the model to be able to process these input data, it needs to be embedded in a higher dimensional space, typically using position coding to represent the sequential information of the time series data. The calculation formula of the position code is as follows:

(13);

(14);

Where t is the time step, i is the dimension index, d is the embedded dimension, i.e., the vector, and PE represents the position code. By adding position coding, the relative relationship between data points in the time series can be captured. The core of the transducer is the Self-Attention mechanism (Self-Attention). The self-attention mechanism allows the model to efficiently learn long-range dependencies of data while processing the data at each point in time, taking into account the data at other times in the global time horizon. The calculation formula of the self-attention mechanism is as follows:

(15);

Where Q, K, V represents the matrix of queries, keys, and values, respectively, d _k is the dimension of the key vector. Specifically, query Q and key K are each derived from an embedded representation of the input data, and value V represents characteristic information of the input data. By means of a self-attention mechanism, the model can assign a weight to the input data at each point in time, which weight reflects the correlation between the current point in time and other points in time. These weights can help the model learn important patterns and dependencies in the data, revealing potential anomalies in the data.

In step S6, the abnormality score of each time step may be further calculated based on the predicted value of the prediction model. Assume that at time step t, a first anomaly A (t) of the model output is calculated by the following formula:

(16);

wherein A (t) is the first anomaly of time step t, X _i2 (t) represents the input data/detection value of the i2 nd sensor at time step t, D2 is the total number of sensors for predicting the predicted value of the model at time step t. The first abnormality degree score a (t) indicates a difference between the data at the current time point and the model predictive value, and the greater the difference is, the higher the abnormality degree at the time point is. When the first degree of abnormality exceeds a set first threshold, the time step t is considered to be abnormal, and an alarm mechanism can be triggered. In order to improve the accuracy and sensitivity of anomaly detection, the setting of the first threshold P1 is critical. Typically, the first threshold P1 is determined by the statistical properties of the historical data, the traffic demand and the security criteria. Assuming that a _max represents the maximum anomaly value in the historical data, which is a safety factor, the first threshold P1 can be set by the following formula:

(17);

When the first abnormality a (T) of the real-time data exceeds the threshold T, it is determined that abnormality has occurred at that point in time, an alarm mechanism is started, and step S9 is executed.

The alarm mechanism is a further key step in the present invention to achieve security and risk control. The method has the main functions of responding and processing the abnormality detected in the oil well construction process in real time according to the output result of the dynamic abnormality detection module and by combining the comprehensive analysis of multi-mode data fusion. Through scientific evaluation and reasonable alarm strategies, the alarm mechanism can accurately transmit abnormal information to related personnel and provide corresponding processing suggestions. Not only improves the safety in the construction process, but also reduces the interference caused by false alarm or missing alarm, and provides reliable technical support for the stable operation of oil well construction.

The background of alarm mechanisms stems from complex anomalies that may exist during well construction. These anomalies are often bursty and complex, and may involve coordinated changes in the multidimensional data. If there is no timely alert mechanism, the anomaly may cause a major security incident without being noticed. Therefore, real-time performance, accuracy and adaptability are required to be considered, so that an alarm can be triggered rapidly under any abnormal condition, and constructors are prompted to take necessary measures.

More specifically, in step S9, sensor data within a time window is givenFirst, a second degree of anomaly a (X _i3) of each piece of data is calculated:

(18);

Wherein X _i3 is the detection value of the i3 rd time window, In order to be able to predict the value,Is the standard deviation of the historical data. However, the method of calculating the second degree of abnormality is not limited thereto, and may be the same as or different from the first degree of abnormality.

Average anomaly degree of current time windowExpressed as:

(19);

Where l is denoted as the number of data stripes under the current time window i 3.

The security score S is calculated by:

(20);

Wherein, the The weight coefficient expressed as the influence of the second anomaly on the safety score is determined by the historical data and expert experience. The safety score S ranges from 0 to 100, with higher values indicating safer construction conditions. When the safety score S is lower than a preset threshold (for example, 80 minutes), the system judges that the construction state has potential safety hazards and triggers an alarm.

The alarm mechanism not only needs to judge whether to trigger an alarm, but also needs to set an alarm level according to the severity of the abnormality and provide corresponding processing advice. The alarm level is classified based on the level of the security score and the type and severity of the anomaly output by the dynamic anomaly detection module. The hierarchical alarm strategy comprises three levels, wherein the second threshold is determined as a plurality of threshold intervals, namely, a low-level alarm is determined, when the safety score is slightly lower than a third threshold interval (for example, the score is 70 and less than or equal to S < 80), the system triggers the low-level alarm to prompt constructors to observe data trend and conduct preventive inspection. Medium level alarm when the fourth threshold interval of the safety score is further reduced (e.g., 50 points +.s < 70 points), the system triggers a medium level alarm suggesting a pause in construction and detailed inspection of the outlier data points. High level alert when the fifth threshold interval of the safety score is very low (e.g., S < 50 points), the high level alert is triggered, construction is stopped immediately and the emergency response procedure is initiated. The corresponding second threshold decision formula for the alert level is:

(21);

where L is the alarm level, and the value is 1 (low), 2 (medium), or 3 (high).

After the alarm level is determined, the alarm information can be sent to related personnel through multiple channels, including short messages, mails, application pushing and the like. The content of the alert information includes the detailed location, type, security score, anomaly cause analysis, and suggested countermeasures of the anomaly data. For example, when the pressure value of a certain sensor deviates significantly from the normal range, the alarm information may include the number of the sensor, the actual value of the pressure data compared to the predicted value, and possible causes (such as sensor failure or actual pressure abnormality). Meanwhile, the visual alarm information display interface can be used for displaying an abnormal data trend graph, a safety grading change curve, alarm grade distribution and the like, so that constructors can be helped to quickly know the construction state and make a correct decision. In order to improve the adaptability of the alarm mechanism, the threshold value is dynamically adjusted. The alarm threshold and classification strategy may be updated in real-time as the construction environment and monitoring data change. For example, during certain critical construction phases (e.g., high pressure testing or deep drilling), the system may dynamically decrease the relevant thresholds for safety scores, increasing the sensitivity of alarms, and thus, finding potential problems earlier.

The dynamically adjusted core algorithm is based on statistical analysis of historical data. The system can dynamically adjust the threshold and the weight by using a Bayesian updating method by analyzing the correlation between the historical alarm record and the actual event. For example, for a particular construction phase, the updated alarm threshold is calculated by:

(22);

Wherein T is an initial threshold or a threshold interval boundary, the correlation coefficient R represents the matching degree of the alarm event and the real abnormality in the historical data, and the historical false alarm rate Wb represents the past alarm error proportion. The correlation coefficient R expression is as follows:

(23);

Wherein, the Indicating the probability of actually occurring an abnormality after triggering an alarm, indicating the reliability of the alarm; the probability of giving an alarm when an abnormality occurs, i.e., the proportion of historically effective alarms, P (E) is the prior probability of the abnormality, i.e., the overall frequency of occurrence of the abnormality, and P (A) is the overall frequency of occurrence of the alarm.

Through the safety score calculation, the grading alarm strategy and the dynamic adjustment function, the efficient response to the abnormal event in the oil well construction process is realized. Not only can the abnormality be found in time, but also a hierarchical alarm and a handling suggestion can be provided according to the severity of the abnormality. Through a scientific algorithm and a multi-channel notification mechanism, the safety risk can be effectively reduced, and the overall safety and management efficiency of oil well construction are improved. The self-adaptive adjustment function of the alarm mechanism enables the alarm mechanism to flexibly adapt to the safety monitoring requirements of different construction stages, and the practicability and reliability of safety detection/monitoring are further enhanced.

The second aspect of the present invention also provides a monitoring system for implementing the above oil well construction safety monitoring method, as shown in fig. 2, which includes an acquisition module 1, a preprocessing module 2, a feature extraction module 3, a fusion module 4, an anomaly detection module 5 and an alarm module 6.

The system comprises an acquisition module 1, a preprocessing module 2 and a control module, wherein the acquisition module 1 is used for acquiring monitoring data of oil well construction, and the preprocessing module 2 is used for preprocessing the monitoring data. The feature extraction module 3 is used for extracting features from the monitored data. The fusion module 4 is used for fusing the characteristics and the monitoring data to obtain fusion characteristics. The anomaly detection module 5 is configured to obtain a predicted value based on the prediction model, obtain a first anomaly degree according to the predicted value and the detected value, and call the alarm module 6 if the first anomaly degree exceeds a first threshold value. The alarm module 6 is used for obtaining a second degree of abnormality and a safety score according to the predicted value and the detected value, and carrying out abnormal alarm according to the safety score and the second threshold value.

The detection system further comprises a training module 7, wherein the training module 7 is used for constructing the training set according to the fusion characteristics, and training the training set based on a machine learning method to obtain a prediction model.

A third aspect of the present invention provides a monitoring apparatus, as shown in fig. 2, comprising a processor and a memory, the memory storing code/program for implementing the above-mentioned security monitoring method, the code/program executing when processed by the processor.

The invention realizes the safety monitoring and risk early warning of the oil well construction data through four core steps of data acquisition and preprocessing, multi-mode data fusion, dynamic anomaly detection and alarm mechanism. And carrying out real-time anomaly identification according to the fused data, triggering an alarm mechanism and timely early warning potential safety risks. The invention provides an efficient and accurate safety monitoring means for oil well construction by integrating various data sources and intelligent algorithms, and obviously improves the safety and reliability in the construction process. The problems of complex data sources, high real-time requirements, insufficient abnormality detection precision and the like are solved.

In the oil well construction process, the complexity and the variability of real-time data require high real-time performance and adaptability. The invention realizes accurate monitoring of real-time data through a dynamic anomaly detection algorithm. Specifically, the security of the overall data is quantified by calculating a security score within each time window. When the security score is below a set threshold, an alarm is triggered and the location, characteristics and possible causes of the anomaly data are marked. The alarm mechanism also supports multi-channel notification, including short messages, mails and application pushing, so that constructors can acquire abnormal information and take corresponding measures in the shortest time.

The invention optimizes the whole flow from data acquisition to anomaly detection to alarm response through the streaming data processing architecture, so that the delay of the whole process is controlled in the millisecond range. The real-time performance is particularly critical in high-risk environments of oil well construction, and potential loss or danger caused by abnormal failure to be found in time can be greatly reduced. In addition, the design of the invention fully considers the stage characteristics of the construction process. The monitoring of data at different stages of oil well construction is different in importance, for example, the state of equipment start-up is more focused in the early stage, and the monitoring of core indexes such as pressure, flow and the like is more emphasized in the construction period. According to the invention, by dynamically adjusting the detection strategy and combining the scene specific rules and the intelligent algorithm, the system can flexibly adapt to different requirements in the construction process. The self-adaption not only improves the practical application value of the system, but also lays a foundation for wide popularization in complex industrial environments.

In summary, the invention solves a plurality of problems in oil well construction data safety monitoring through the design of the intelligent data safety monitoring method and the alarm system. By combining dynamic anomaly detection, multi-mode data fusion and intelligent alarm, a real-time, efficient and highly-adaptive safety monitoring system is constructed, and comprehensive safety guarantee is provided for oil well construction. Meanwhile, the design concept and the technical architecture of the invention also provide an important reference direction for subsequent intelligent processing and abnormal prevention technical research.

The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. The safety monitoring method for the oil well construction is characterized by comprising the following steps of:

collecting monitoring data of oil well construction;

extracting features from the monitoring data, and fusing the features and the monitoring data to obtain fused features, wherein the fused features comprise any one index or combination of the following indexes, namely data source features, feature level features and decision level features;

Predicting the detection value of the fusion characteristic through a prediction model based on machine learning to obtain a prediction value;

calculating a first anomaly degree according to the predicted value and the detected value;

judging whether the first abnormality degree exceeds a first threshold value;

if yes, obtaining a second anomaly degree and a safety score according to the predicted value and the detected value, wherein the second anomaly degree is expressed as:

;

Wherein A (Xi 3) is the second degree of anomaly, xi3 is the detection value of the i3 th time window, In order to be able to predict the value,Standard deviation of historical data;

The security evaluation S is expressed as:

;

Wherein, the A weight coefficient expressed as the influence of the second anomaly on the security score, and l expressed as the number of data bars under the current time window i 3;

and carrying out abnormal alarm according to the safety score and the second threshold value.

2. The method of claim 1, wherein the extracted features include any one or a combination of temperature, pressure, flow, humidity, rate of pressure change, temperature gradient, and flow fluctuation amplitude;

3. The safety monitoring method according to claim 2, wherein the rate of pressure change is expressed as:

;

temperature gradient Expressed as:

;

Flow fluctuation amplitude Expressed as:

;

the data source signature F _fusion is represented as:

;

feature level feature X _fusion is represented as:

=;

the decision level feature Y _fusion is represented as:

;

4. The security monitoring method of claim 1, wherein the weights are updated based on a bayesian formulation expressed as:

;

Where w is denoted as a weight selected from the group consisting of the weight of the data source, the weight of the feature vector and the weight of the decision output, For the posterior probability of the weights in the case of data D,For the likelihood of the data D under weight,For the a priori distribution of weights, P (D) is the total probability of data D.

5. The method of claim 1, wherein the machine learning method comprises a support vector machine, a decision tree, or a neural network,

The neural network includes a transducer.

6. The method of claim 1, wherein the first anomaly is calculated by:

;

wherein A (t) is the first anomaly of time step t, X _i2 (t) represents the input data/detection value of the i2 nd sensor at time step t, (T) is a predicted value of the prediction model at time step t, and d2 is the total number of sensors;

The first threshold P1 is defined as:

;

Wherein, the Represents the maximum outlier in the history data,Is a safety factor.

7. The safety monitoring method according to claim 1, wherein the level determination of the abnormality alarm is:

;

where L is denoted as the level of the alarm.

8. The method of claim 1, further comprising a method of updating a boundary of a threshold or threshold interval, the updated threshold or threshold intervalExpressed as:

;

The correlation coefficient R expression is as follows:

;

Wherein, the Indicating the probability of actually occurring an abnormality after triggering an alarm, indicating the reliability of the alarm; The probability of an alarm being given when an abnormality occurs, P (E) is the prior probability of the abnormality, and P (A) is the overall occurrence frequency of the alarm.

9. A monitoring system for implementing the safety monitoring method according to any one of claims 1-8, the monitoring system comprising an acquisition module, a feature extraction module, a fusion module, an anomaly detection module, and an alarm module;