Fault prediction method and device for reducing misoperation probability of digital gas relay and medium

Abstract

The invention relates to the field of transformer relay protection, in particular to a fault prediction method and device for reducing the misoperation probability of a digital gas relay and a medium. The method comprises the following steps: firstly collecting data, then performing cleaning processing, eliminating interference by using multiple filtering modes, analyzing filtered data based on a 3 sigma rule and an isolated forest algorithm, identifying abnormal data points, eliminating the abnormal data points, dividing signal data into a high-confidence region and a low-confidence region based on a preset fault judgment threshold value, and performing fault diagnosis on the signal data according to the high-confidence region and the low-confidence region. And for the signal data of the high-confidence region, generating a preliminary fault judgment result directly based on a rule judgment method, and for the signal data of the low-confidence region, inputting the signal data into a deep learning classification model, analyzing the low-confidence signal data by using the pre-trained deep learning classification model, and outputting a fault classification result. The detection data is processed, so that the false alarm probability is reduced.

Description

Technical Field

[0001] The present invention relates to the field of transformer relay protection, and in particular to a fault prediction method, device and medium for reducing the malfunction probability of a digital gas relay for a transformer. Background Art

[0002] With the rapid development of digitalization and intelligence of power systems, gas relays, as important protection devices for transformers and other oil-immersed power equipment, have a direct impact on the operating safety of the equipment with their reliability and accuracy. Gas relays mainly monitor the gas release rate and related signals generated during the operation of the equipment to determine whether there are potential faults. However, in the existing technology, the false operation rate of digital gas relays is relatively high, which seriously affects the operating efficiency and safety of power equipment, especially under complex working conditions.

[0003] At present, most gas relays use simple signal analysis and fixed threshold judgment methods to determine whether to trigger protection actions. The traditional method can effectively respond to some common faults, but it has significant limitations. First, the existing technology has a relatively simple processing method for raw signal data, which is difficult to deal with data interference in complex environments. For example, high-frequency noise, nonlinear interference and signal fluctuations can lead to misjudgment. The misoperation rate of gas relays increases significantly under complex working conditions of high temperature, high humidity or strong electromagnetic interference. Secondly, the existing methods are not capable of identifying abnormal data, and cannot accurately remove noise and abnormal points in the sampled data, which can easily lead to inaccurate protection actions due to data quality problems. In recent years, some studies have introduced data processing and intelligent algorithms to analyze gas relay signals, and tried to improve the accuracy of signal judgment through machine learning and statistical methods. However, the following problems are exposed in practical applications: On the one hand, most methods fail to fully combine the characteristics of high confidence and low confidence areas, and it is difficult to effectively balance the real-time nature of data and the accuracy of judgment in a variety of complex environments; on the other hand, the existing technology lacks a systematic data preprocessing strategy and fails to optimize the entire process from data collection, cleaning to judgment. In addition, current intelligent methods are inefficient in processing large-scale real-time data and cannot quickly adapt to the dynamic changes of different working conditions, resulting in delayed protection actions or missed judgments. In summary, the existing technology has significant deficiencies in the accuracy of data processing under complex working conditions, the intelligent level of signal analysis, and the real-time and reliability of protection actions, which seriously affects the actual application effect of digital gas relays. A systematic and efficient solution is urgently needed to improve its reliability and accuracy under complex working conditions. Summary of the invention

[0004] In view of the defects of the prior art, the present invention provides a fault prediction method, device and medium for reducing the probability of malfunction of a digital gas relay, and reduces the probability of false alarm and improves the accuracy of fault prediction by processing the detection data.

[0005] In order to solve the technical problem, the technical solution adopted by the present invention is: a fault prediction method for reducing the probability of malfunction of a digital gas relay, comprising the following steps: S1. Collect real-time signal data of equipment operation, including gas release rate data, temperature data, humidity data and electromagnetic signal strength data, through the sensor module of the digital gas relay to form a set of original signal data containing multiple dimensions; S2. Clean up the collected original signal data set, use adaptive filtering method to suppress high-frequency noise, use signal smoothing algorithm to eliminate low-amplitude fluctuations, and reduce nonlinear interference in the signal by segmented processing; S3, analyzing the original signal data set after data cleaning based on the 3σ rule and the isolation forest algorithm, identifying abnormal data points in the original signal data set, removing the abnormal data points from the original signal data set, and generating the original signal data set after abnormality removal; S4. Analyze the original signal data set after the abnormality is eliminated by the rule determination method, and divide the signal data into a high confidence area and a low confidence area based on the preset fault determination threshold. The signal data in the high confidence area is the normal operating state data that complies with the rules, and the signal data in the low confidence area is the data that is difficult to clearly classify; S5. For the signal data in the high confidence area, a preliminary fault determination result is directly generated based on the rule determination method. For the signal data in the low confidence area, the signal data is input into the deep learning classification model, and the low confidence signal data is analyzed using the pre-trained deep learning classification model to output the fault classification result. S6. Fusing the preliminary fault judgment result with the fault classification result to generate a final fault judgment result; S7. According to the final fault judgment result and the actual operating status of the gas relay, record the accuracy of the judgment, dynamically adjust the threshold parameters of the rule judgment method and the classification weight of the deep learning classification model, and continuously optimize the high confidence area recognition and classification judgment strategy; S8. The final fault judgment result is transmitted to the control module of the gas relay as a basis for triggering the protection action.

[0006] Furthermore, step S1 is specifically as follows: S11. Collect signal data X through the sensor module of the digital gas relay, obtain gas release rate data, temperature data, humidity data and electromagnetic signal strength data respectively, and record the sampling time of each signal data , , in It represents the value of the i-th signal at the j-th sampling point, where m is the signal type and n is the number of sampling points; S12, normalize the signal data X to generate a normalized signal matrix ; S13, according to the timestamp of real-time signal data The normalized signal matrix Perform time series segmentation, each segment includes data of k consecutive sampling points, forming a segment matrix set : , Where q is the segment index, Indicates the index of the sampling points contained in the segment. Indicates the total number of segments into which the signal data is segmented; S14. Store the p segment matrices as the original signal data set D: .

[0007] Furthermore, step S2 is specifically as follows: S21, using an adaptive filtering method to suppress high-frequency noise in the original signal data set, and obtaining a filtered signal matrix ; S22, the signal matrix after filtering Perform signal smoothing to obtain the smoothed signal matrix : , in For the matrix The signal value of row i and column j, is the smoothing window width, is the signal value within the window; S23, signal matrix after signal smoothing Perform nonlinear interference detection by calculating the nonlinear interference coefficient To judge the matrix Is there significant nonlinear interference? , in For the matrix The mean of the i-th row, m is the number of signal types, k is the number of matrix columns, if Exceeding the preset threshold , then it is considered that there is nonlinear interference; S24, signal matrix with nonlinear interference Generate signal matrix by using segmented reconstruction algorithm to remove nonlinear interference , for a matrix without nonlinear interference, directly let ; S25, all the signal matrices after removing interference Reassemble into a cleaned signal data set .

[0008] Furthermore, step S3 is specifically as follows: S31, the signal data set after cleaning Each signal matrix Perform statistical analysis and calculate the mean of each signal and standard deviation ; S32, based on the 3σ rule, for each signal value in the signal matrix Perform preliminary anomaly detection to determine whether the conditions are met: ; If the above conditions are met, the signal value Mark as abnormal data points; S33, the signal matrix after preliminary anomaly detection The isolation forest algorithm is applied for further anomaly detection, and the isolation forest model IF is constructed. Input the isolation forest model IF and calculate the anomaly score for each sampling point : , in represents the multidimensional anomaly score of the jth sampling point, T is the set of decision trees in the isolation forest, are any two nodes in the decision tree, , Represents the signal matrix The sampling value of the sample corresponding to the internal nodes t and t' at the jth sampling point, is the Euclidean distance between the samples corresponding to nodes t and t' at the jth sampling point, Represents the difference in path depth between nodes t and t' in the isolation forest model, which is used to measure the degree of isolation between samples. is a positive number; like ,in is the abnormal score threshold, then all signal values ​​of the corresponding sampling point j Mark as abnormal data points; S34, all abnormal data points detected by the 3σ rule and the isolation forest algorithm are removed from the signal matrix Eliminate the abnormalities and generate a signal matrix after eliminating the abnormalities ; S35, signal matrix after all abnormalities are eliminated Recombined to form the original signal data set after the abnormality is removed .

[0009] Furthermore, step S4 is specifically as follows: S41, the original signal data set after the abnormality is eliminated Each signal matrix in Perform rule determination analysis for each signal value Based on the preset fault judgment threshold , determine whether the high confidence condition is met: ; in, is the mean value of the ith signal, is the fault judgment threshold preset for the i-th signal; if the above conditions are met, Marked as a high confidence signal value and recorded in the high confidence area set Otherwise, Marked as low confidence signal value and recorded in the low confidence area set ; S42, for signal matrix All signal values ​​of form a high confidence region matrix and the low confidence region matrix : , in Represents the signal matrix The signal value of the high confidence area that meets the rule judgment, Represents the signal matrix The signal value of the low confidence area that does not meet the rule judgment; S43, the high confidence region matrix of each signal matrix and the low confidence region matrix Store them separately according to the matrix index q to form a high confidence area signal data set and low confidence area signal data set : ; Where p is the total number of segments of the signal matrix.

[0010] Furthermore, step S5 is specifically as follows: S51. Signal data set for high confidence area Each high confidence region matrix in , calculate the preliminary fault judgment result based on the rule judgment method: , in A set of signal values ​​representing compliance with the rule determination; like , then the high confidence area is judged to be in normal operation state, otherwise it is judged to be in potential failure state. The normal state determination ratio threshold for the high confidence area; S52, low confidence area signal data set Each low confidence region matrix Input the deep learning classification model, use the deep learning classification model to classify the low confidence area matrix, and generate the fault classification results .

[0011] Furthermore, step S6 is specifically as follows: S61. Obtaining a set of preliminary fault determination results in a high confidence area And the fault classification result set of low confidence area , respectively calculate the comprehensive judgment result weights of the high confidence area and the weight of the comprehensive classification result of the low confidence area : , in, is the number of signal values ​​in the high confidence region, is the number of signal values ​​in the low confidence area; S62, based on the comprehensive judgment result weight and the weight of the comprehensive classification results of the low confidence area The preliminary fault judgment result set in the high confidence area and the fault classification result set in the low confidence area are weighted and fused to generate a comprehensive fault score. : ; in, is the fault score of the high confidence area, which is directly calculated based on the rule judgment result. If it is a potential fault, ,otherwise ; Output the fault classification probability distribution for the low confidence area; S63. Setting comprehensive fault scoring threshold The comprehensive fault score for each segment matrix q Make a judgment, if , then the segment is considered to correspond to the fault state, and the final fault judgment result set is generated : , in is a dynamic adjustment coefficient used to comprehensively evaluate the global influence of the segment. S64, record the individual judgment results of the high confidence area and the low confidence area, and combine them with the comprehensive fault judgment result set Output the final report, including the operating status of each segment, potential fault location and fault category information.

[0012] Furthermore, step S8 is specifically as follows: S81. Obtaining a comprehensive fault judgment result set , encode the state information corresponding to each segment index q in the set and generate the fault state encoding matrix E: , in, Indicates the fault status code of segment q, 1 indicates fault status, and 0 indicates normal status; S82, the fault state encoding matrix E and the segment matrix index q and the corresponding timestamp Correlation, generating fault trigger signal sequence ; S83, the fault trigger signal sequence Input the gas relay control module and analyze the fault status information and corresponding time , to trigger the decision of protection action; when When a fault occurs, the control module immediately performs protection actions, including power-off operation, alarm triggering or data storage; S84, collect the comprehensive fault judgment results and fault trigger signal sequence The analysis results are stored as historical operation records; S85: Adjust the comprehensive fault score threshold according to the operation feedback of the gas relay control module Or dynamic adjustment coefficient , optimize the next step of fault determination process.

[0013] The present invention also discloses a fault prediction device for reducing the probability of malfunction of a digital gas relay, comprising a processor and a memory storing program instructions, wherein the processor is configured to execute the fault prediction method for reducing the probability of malfunction of a digital gas relay as described above when running the program instructions.

[0014] The present invention also discloses a storage medium storing program instructions, which, when running, execute the fault prediction method for reducing the probability of malfunction of the digital gas relay as described above.

[0015] Beneficial effects of the present invention: The present invention divides the signal data set after anomalies are eliminated into high-confidence areas and low-confidence areas, processes the signals in the high-confidence area using a rule judgment method, and uses a deep learning classification model to analyze the signals in the low-confidence area, thereby realizing hierarchical processing of data. Compared with the method of uniformly analyzing all data with a single algorithm in the prior art, the present invention can select the most suitable processing method according to the confidence of the data, reduce the calculation complexity while ensuring the judgment accuracy, and improve the judgment efficiency. Under complex working conditions, the false operation rate caused by the uncertainty of the signals in the low-confidence area is effectively reduced, and the reliability and adaptability of the gas relay are significantly enhanced. Experiments show that after adopting this method, the fault judgment accuracy in the high-confidence area is improved by 12%, and the overall calculation time is reduced by 35%.

[0016] The present invention proposes a systematic data cleaning and fault result fusion algorithm, which first eliminates high-frequency noise and nonlinear interference in the signal through adaptive filtering and piecewise smoothing, and then removes abnormal data through the 3σ rule and the isolation forest algorithm. Finally, based on the judgment results of high confidence and low confidence, a weighted fusion method is used to generate a comprehensive fault score and obtain the final fault judgment result. Compared with the single anomaly detection and classification method in the prior art, the method dynamically optimizes the algorithm parameters in each step to adapt to the variable working conditions, thereby ensuring the robustness of data processing. Experimental results show that the detection accuracy of the anomaly removal algorithm of the present invention is improved by 15% in a noisy environment, and the accuracy of the final fault classification is improved by about 10%.

[0017] The present invention designs a comprehensive fault scoring model, introduces dynamic adjustment coefficients and comprehensive weights, can dynamically optimize the scoring mechanism according to real-time data, uniformly evaluate the global data and adjust the judgment threshold. Compared with the fixed threshold protection judgment strategy in the prior art, the method of the present invention is more flexible, can maintain consistent judgment standards under different operating states and data distributions, and avoids protection action errors caused by insufficient adaptability of a single fixed threshold. Tests show that under multiple working conditions, the dynamic adjustment mechanism of the present invention can reduce the misjudgment rate by 20% and the missed judgment rate by 15%, significantly improving the reliability of the gas relay protection action. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A flow chart of a data preprocessing method for reducing the probability of malfunction of a digital gas relay proposed by the present invention; Figure 2 Schematic diagram of the device described in Example 2. DETAILED DESCRIPTION

[0019] The present invention will be further described below in conjunction with the accompanying drawings and specific implementations.

[0020] Example 1 This embodiment discloses a fault detection method for reducing the probability of malfunction of a digital gas relay. Figure 1 As shown, the following steps are included: S1. Collect real-time signal data of equipment operation, including gas release rate data, temperature data, humidity data and electromagnetic signal strength data, through the sensor module of the digital gas relay to form a set of original signal data containing multiple dimensions; S2. Clean up the collected original signal data set, use adaptive filtering method to suppress high-frequency noise, use signal smoothing algorithm to eliminate low-amplitude fluctuations, and reduce nonlinear interference in the signal by segmented processing; S3, analyzing the original signal data set after data cleaning based on the 3σ rule and the isolation forest algorithm, identifying abnormal data points in the original signal data set, removing the abnormal data points from the original signal data set, and generating the original signal data set after abnormality removal; S4. Analyze the original signal data set after the abnormality is eliminated by the rule determination method, and divide the signal data into a high confidence area and a low confidence area based on the preset fault determination threshold. The signal data in the high confidence area is the normal operating state data that complies with the rules, and the signal data in the low confidence area is the data that is difficult to clearly classify; S5. For the signal data in the high confidence area, a preliminary fault determination result is directly generated based on the rule determination method. For the signal data in the low confidence area, the signal data is input into the deep learning classification model, and the low confidence signal data is analyzed using the pre-trained deep learning classification model to output the fault classification result. S6. Fusing the preliminary fault judgment result with the fault classification result to generate a final fault judgment result; S7. According to the final fault judgment result and the actual operating status of the gas relay, record the accuracy of the judgment, dynamically adjust the threshold parameters of the rule judgment method and the classification weight of the deep learning classification model, and continuously optimize the high confidence area recognition and classification judgment strategy; S8. The final fault judgment result is transmitted to the control module of the gas relay as a basis for triggering the protection action.

[0021] In this embodiment, step S1 is specifically as follows: S11. Collect signal data X through the sensor module of the digital gas relay, obtain gas release rate data, temperature data, humidity data and electromagnetic signal strength data respectively, and record the sampling time of each signal data : , in It represents the value of the i-th signal at the j-th sampling point, where m is the signal type and n is the number of sampling points; S12, normalize the signal data X to generate a normalized signal matrix ; S13, according to the timestamp of real-time signal data The normalized signal matrix Perform time series segmentation, each segment includes data of k consecutive sampling points, forming a segment matrix set : , Where q is the segment index, Indicates the index of the sampling points contained in the segment. Indicates the total number of segments into which the signal data is segmented. Indicates rounding down; S14. Store the p segment matrices as the original signal data set D: .

[0022] In this embodiment, step S2 is specifically as follows: S21, using an adaptive filtering method to filter each segment matrix in the original signal data set D The high-frequency noise in the filter is suppressed to obtain the filtered signal matrix ; S22, the signal matrix after filtering Perform signal smoothing to obtain the smoothed signal matrix : , in For the matrix The signal value of row i and column j, is the smoothing window width, is the signal value within the window; S23, signal matrix after signal smoothing Perform nonlinear interference detection by calculating the nonlinear interference coefficient To judge the matrix Is there significant nonlinear interference? , in For the matrix The mean of the i-th row, m is the number of signal types, k is the number of matrix columns, if Exceeding the preset threshold , then it is considered that there is nonlinear interference; S24, signal matrix with nonlinear interference Generate signal matrix by using segmented reconstruction algorithm to remove nonlinear interference , for a matrix without nonlinear interference, directly let ; S25, all the signal matrices after removing interference Reassemble into a cleaned signal data set .

[0023] In this embodiment, step S3 is specifically as follows: S31, the signal data set after cleaning Each signal matrix Perform statistical analysis and calculate the mean of each signal and standard deviation ; S32, based on the 3σ rule, for each signal value in the signal matrix Perform preliminary anomaly detection to determine whether the conditions are met: ; If the above conditions are met, the signal value Mark as abnormal data points; S33, the signal matrix after preliminary anomaly detection The isolation forest algorithm is applied for further anomaly detection, and the isolation forest model IF is constructed. Input the isolation forest model IF and calculate the anomaly score for each sampling point : , in represents the multidimensional anomaly score of the jth sampling point, T is the set of decision trees in the isolation forest, are any two nodes in the decision tree, , Represents the signal matrix The sampling value of the sample corresponding to the internal nodes t and t' at the jth sampling point, is the Euclidean distance between the samples corresponding to nodes t and t' at the jth sampling point, Represents the difference in path depth between nodes t and t' in the isolation forest model, which is used to measure the degree of isolation between samples. is a positive number; like ,in is the abnormal score threshold, then all signal values ​​of the corresponding sampling point j Mark as abnormal data points; S34, all abnormal data points detected by the 3σ rule and the isolation forest algorithm are removed from the signal matrix Eliminate the abnormalities and generate a signal matrix after eliminating the abnormalities ; S35, signal matrix after all abnormalities are eliminated Recombined to form the original signal data set after the abnormality is removed .

[0024] In this embodiment, step S4 is specifically as follows: S41, the original signal data set after the abnormality is eliminated Each signal matrix in Perform rule determination analysis for each signal value Based on the preset fault judgment threshold , determine whether the high confidence condition is met: ; in, is the mean value of the ith signal, is the fault judgment threshold preset for the i-th signal; if the above conditions are met, Marked as a high confidence signal value and recorded in the high confidence area set Otherwise, Marked as low confidence signal value and recorded in the low confidence area set ; S42, for signal matrix All signal values ​​of form a high confidence region matrix and the low confidence region matrix : , in Represents the signal matrix The signal value of the high confidence area that meets the rule judgment, Represents the signal matrix The signal value of the low confidence area that does not meet the rule judgment; S43, the high confidence region matrix of each signal matrix and the low confidence region matrix Store them separately according to the matrix index q to form a high confidence area signal data set and low confidence area signal data set : ; Where p is the total number of segments of the signal matrix.

[0025] In this embodiment, step S5 is specifically as follows: S51. Signal data set for high confidence area Each high confidence region matrix in , calculate the preliminary fault judgment result based on the rule judgment method: , in A set of signal values ​​representing compliance with the rule determination; like , then the high confidence area is judged to be in normal operation state, otherwise it is judged to be in potential failure state. The normal state determination ratio threshold for the high confidence area; S52, low confidence area signal data set Each low confidence region matrix Input the deep learning classification model, use the deep learning classification model to classify the low confidence area matrix, and generate the fault classification results .

[0026] In this embodiment, step S6 is specifically as follows: S61. Obtaining a set of preliminary fault determination results in a high confidence area And the fault classification result set of low confidence area , respectively calculate the comprehensive judgment result weights of the high confidence area and the weight of the comprehensive classification results of the low confidence area : , in, is the number of signal values ​​in the high confidence region, is the number of signal values ​​in the low confidence area; S62, based on the comprehensive judgment result weight and the weight of the comprehensive classification results of the low confidence area The preliminary fault judgment result set in the high confidence area and the fault classification result set in the low confidence area are weighted and fused to generate a comprehensive fault score. : ; in, is the fault score of the high confidence area, which is directly calculated based on the rule judgment result. If it is a potential fault, ,otherwise ; Output the fault classification probability distribution for the low confidence area; S63. Setting comprehensive fault scoring threshold The comprehensive fault score for each segment matrix q Make a judgment, if , then the segment is considered to correspond to the fault state, and the final fault judgment result set is generated : , in is a dynamic adjustment coefficient used to comprehensively evaluate the global influence of the segment. S64, record the individual judgment results of the high confidence area and the low confidence area, and combine them with the comprehensive fault judgment result set Output the final report, including the operating status of each segment, potential fault location and fault category information.

[0027] In this embodiment, step S8 is specifically as follows: S81. Obtaining a comprehensive fault judgment result set , encode the state information corresponding to each segment index q in the set and generate the fault state encoding matrix E: , in, Indicates the fault status code of segment q, 1 indicates fault status, and 0 indicates normal status; S82, the fault state encoding matrix E and the segment matrix index q and the corresponding timestamp Correlation, generating fault trigger signal sequence ; S83, the fault trigger signal sequence Input the gas relay control module and analyze the fault status information and corresponding time , to trigger the decision of protection action; when When a fault occurs, the control module immediately performs protection actions, including power-off operation, alarm triggering or data storage; S84, collect the comprehensive fault judgment results and fault trigger signal sequence The analysis results are stored as historical operation records; S85: Adjust the comprehensive fault score threshold according to the operation feedback of the gas relay control module Or dynamic adjustment coefficient , optimize the next step of fault determination process.

[0028] Example 2 Combination Figure 2 As shown, the embodiment of the present disclosure provides a fault prediction device for reducing the probability of malfunction of a digital gas relay, including a processor and a memory. Optionally, the device may also include a communication interface and a bus. The processor 304, the communication interface, and the memory may communicate with each other through the bus. The communication interface may be used for information transmission. The processor may call the logic instructions in the memory to execute the fault prediction method for reducing the probability of malfunction of a digital gas relay of embodiment 1.

[0029] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.

[0030] The memory, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as program instructions / modules corresponding to the method in the embodiment of the present disclosure. The processor executes the function application and data processing by running the program instructions / modules stored in the memory, that is, the fault prediction method for reducing the probability of malfunction of the digital gas relay in Example 1 is implemented.

[0031] The memory may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal device, etc. In addition, the memory may include a high-speed random access memory and may also include a non-volatile memory.

[0032] Example 3 The embodiment of the present disclosure provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured to execute the fault prediction method for reducing the probability of malfunction of a digital gas relay as described in Embodiment 1.

[0033] The computer-readable storage medium mentioned above may be a transient computer-readable storage medium or a non-transitory computer-readable storage medium.

[0034] The technical solution of the embodiment of the present disclosure can be embodied in the form of a software product, which is stored in a storage medium and includes one or more instructions for enabling a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the method described in the embodiment of the present disclosure. The aforementioned storage medium may be a non-transient storage medium, including: a USB flash drive, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes, or a transient storage medium.

[0035] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent only possible changes. Unless explicitly required, separate components and functions are optional, and the order of operation may vary. The parts and features of some embodiments may be included in or replace the parts and features of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the scope of protection. As used in the description in the text, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of one or more associated listings. In addition, when used in this application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of stated features, wholes, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or groups thereof. In the absence of further restrictions, the elements defined by the sentence "including one..." do not exclude the presence of other identical elements in the process, method or device including the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.

[0036] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. The technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present disclosure. The technicians may clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here.

Claims

1. A fault prediction method for reducing the probability of malfunction of a digital gas relay, characterized in that: The following steps are involved: S1. Collect real-time signal data of equipment operation, including gas release rate data, temperature data, humidity data and electromagnetic signal strength data, through the sensor module of the digital gas relay to form a set of original signal data containing multiple dimensions; S2. Clean up the collected original signal data set, use adaptive filtering method to suppress high-frequency noise, use signal smoothing algorithm to eliminate low-amplitude fluctuations, and reduce nonlinear interference in the signal by segmented processing; S3, analyzing the original signal data set after data cleaning based on the 3σ rule and the isolation forest algorithm, identifying abnormal data points in the original signal data set, removing the abnormal data points from the original signal data set, and generating the original signal data set after abnormality removal; S4. Analyze the original signal data set after the abnormality is eliminated by the rule determination method, and divide the signal data into a high confidence area and a low confidence area based on the preset fault determination threshold. The signal data in the high confidence area is the normal operating state data that complies with the rules, and the signal data in the low confidence area is the data that is difficult to clearly classify; S5. For the signal data in the high confidence area, a preliminary fault determination result is directly generated based on the rule determination method. For the signal data in the low confidence area, the signal data is input into the deep learning classification model, and the low confidence signal data is analyzed using the pre-trained deep learning classification model to output the fault classification result. S6. Fusing the preliminary fault judgment result with the fault classification result to generate a final fault judgment result; S7. According to the final fault judgment result and the actual operating status of the gas relay, record the accuracy of the judgment, dynamically adjust the threshold parameters of the rule judgment method and the classification weight of the deep learning classification model, and continuously optimize the high confidence area recognition and classification judgment strategy; S8. The final fault judgment result is transmitted to the control module of the gas relay as a basis for triggering a protection action.

2. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 1, characterized in that: Step S1 is specifically as follows: S11. Collect signal data X through the sensor module of the digital gas relay, obtain gas release rate data, temperature data, humidity data and electromagnetic signal strength data respectively, and record the sampling time of each signal data , ， in It represents the value of the i-th signal at the j-th sampling point, where m is the signal type and n is the number of sampling points; S12, normalize the signal data X to generate a normalized signal matrix ; S13, according to the timestamp of real-time signal data The normalized signal matrix Perform time series segmentation, each segment includes data of k consecutive sampling points, forming a segment matrix set : ， Where q is the segment index, Indicates the index of the sampling points contained in the segment. Indicates the total number of segments into which the signal data is segmented; S14. Store the p segment matrices as the original signal data set D: 。 3. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 1, characterized in that: Step S2 is specifically as follows: S21, using an adaptive filtering method to suppress high-frequency noise in the original signal data set, and obtaining a filtered signal matrix ; S22, the signal matrix after filtering Perform signal smoothing to obtain the smoothed signal matrix : ， in For the matrix The signal value of row i and column j, is the smoothing window width, is the signal value within the window; S23, signal matrix after signal smoothing Perform nonlinear interference detection by calculating the nonlinear interference coefficient To judge the matrix Is there significant nonlinear interference? ， in For the matrix The mean of the i-th row, m is the number of signal types, k is the number of matrix columns, if Exceeding the preset threshold , then it is considered that there is nonlinear interference; S24, signal matrix with nonlinear interference Generate signal matrix by using segmented reconstruction algorithm to remove nonlinear interference , for a matrix without nonlinear interference, directly let ; S25, all the signal matrices after removing interference Reassemble into a cleaned signal data set .

4. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 3 is characterized in that: Step S3 is specifically as follows: S31, the signal data set after cleaning Each signal matrix Perform statistical analysis and calculate the mean of each signal and standard deviation ; S32, based on the 3σ rule, each signal value in the signal matrix Perform preliminary anomaly detection to determine whether the conditions are met: ； If the above conditions are met, the signal value Mark as abnormal data points; S33, the signal matrix after preliminary anomaly detection The isolation forest algorithm is applied for further anomaly detection, and the isolation forest model IF is constructed. Input the isolation forest model IF and calculate the anomaly score for each sampling point : ， in represents the multidimensional anomaly score of the jth sampling point, T is the set of decision trees in the isolation forest, are any two nodes in the decision tree, , Represents the signal matrix The sampling value of the sample corresponding to the internal nodes t and t' at the jth sampling point, is the Euclidean distance between the samples corresponding to nodes t and t' at the jth sampling point, Represents the difference in path depth between nodes t and t' in the isolation forest model, which is used to measure the degree of isolation between samples. is a positive number; like ,in is the abnormal score threshold, then all signal values ​​of the corresponding sampling point j Mark as abnormal data points; S34, all abnormal data points detected by the 3σ rule and the isolation forest algorithm are removed from the signal matrix Eliminate the abnormalities and generate a signal matrix after eliminating the abnormalities ; S35, signal matrix after all abnormalities are eliminated Recombined to form the original signal data set after the abnormality is removed .

5. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 4, characterized in that: Step S4 is specifically as follows: S41, the original signal data set after the abnormality is eliminated Each signal matrix in Perform rule determination analysis for each signal value Based on the preset fault judgment threshold , determine whether the high confidence condition is met: ； in, is the mean value of the ith signal, is the fault judgment threshold preset for the i-th signal; if the above conditions are met, Marked as a high confidence signal value and recorded in the high confidence area set Otherwise, Marked as low confidence signal value and recorded in the low confidence area set ; S42, for signal matrix All signal values ​​of form a high confidence region matrix and the low confidence region matrix : ， in Represents the signal matrix The signal value of the high confidence area that meets the rule judgment, Represents the signal matrix The signal value of the low confidence area that does not meet the rule judgment; S43, the high confidence region matrix of each signal matrix and the low confidence region matrix Store them separately according to the matrix index q to form a high confidence area signal data set and low confidence area signal data set : ； Where p is the total number of segments of the signal matrix.

6. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 5, characterized in that: Step S5 is specifically as follows: S51. Signal data set for high confidence area Each high confidence region matrix in , calculate the preliminary fault judgment result based on the rule judgment method: ， in A set of signal values ​​representing compliance with the rule determination; like , then the high confidence area is judged to be in normal operation state, otherwise it is judged to be in potential failure state. The normal state determination ratio threshold for the high confidence area; S52, low confidence area signal data set Each low confidence region matrix Input the deep learning classification model, use the deep learning classification model to classify the low confidence area matrix, and generate the fault classification results .

7. A fault prediction method for reducing malfunction of a digital gas relay according to claim 6, characterized in that: Step S6 is specifically as follows: S61. Obtaining a set of preliminary fault determination results in a high confidence area And the fault classification result set of low confidence area , respectively calculate the comprehensive judgment result weights of the high confidence area and the weight of the comprehensive classification result of the low confidence area : ， in, is the number of signal values ​​in the high confidence region, is the number of signal values ​​in the low confidence area; S62, based on the comprehensive judgment result weight and the weight of the comprehensive classification result of the low confidence area The preliminary fault judgment result set in the high confidence area and the fault classification result set in the low confidence area are weighted and fused to generate a comprehensive fault score. : ； in, is the fault score of the high confidence area, which is directly calculated based on the rule judgment result. If it is a potential fault, ,otherwise ; Output the fault classification probability distribution for the low confidence area; S63. Setting comprehensive fault scoring threshold The comprehensive fault score for each segment matrix q Make a judgment, if , then the segment is considered to correspond to the fault state, and the final fault judgment result set is generated : ， in is a dynamic adjustment coefficient used to comprehensively evaluate the global influence of the segment. S64, record the individual judgment results of the high confidence area and the low confidence area, and combine them with the comprehensive fault judgment result set Output the final report, including the operating status of each segment, potential fault location and fault category information.

8. The fault prediction method for reducing the malfunction probability of a digital gas relay according to claim 7, characterized in that: Get comprehensive fault judgment result set , encode the state information corresponding to each segment index q in the set and generate the fault state encoding matrix E: ， in, Indicates the fault status code of segment q, 1 indicates fault status, and 0 indicates normal status; S82, the fault state encoding matrix E and the segment matrix index q and the corresponding timestamp Correlation, generating fault trigger signal sequence ; S83, the fault trigger signal sequence Input the gas relay control module and analyze the fault status information and corresponding time , to trigger the decision of protection action; when When a fault occurs, the control module immediately performs protection actions, including power-off operation, alarm triggering or data storage; S84, collect the comprehensive fault judgment results and fault trigger signal sequence The analysis results are stored as historical operation records; S85: Adjust the comprehensive fault score threshold according to the operation feedback of the gas relay control module Or dynamic adjustment coefficient , optimize the next step of fault determination process.

9. A fault prediction device for reducing the probability of malfunction of a digital gas relay, comprising a processor and a memory storing program instructions, characterized in that: The processor is configured to execute the fault prediction method for reducing the malfunction probability of a digital gas relay as claimed in any one of claims 1 to 8 when running the program instructions.

10. A storage medium storing program instructions, characterized in that: When the program instructions are executed, the fault prediction method for reducing the probability of malfunction of a digital gas relay as described in any one of claims 1 to 8 is executed.

Images