Method and device for diagnosing faults of a vp tiltmeter

Abstract

This specification provides a method and apparatus for diagnosing faults in VP tiltmeters. The method includes: preprocessing raw fault data from multiple VP tiltmeters and performing CEEMD decomposition to obtain several intrinsic mode functions (EMFs); calculating the distribution entropy of each EMF and using four important network parameters as unknown variables for GOA algorithm optimization; and training the SOM model to obtain a training set S. tr1 SOM clustering label values ​​l of P fault signals p and test set S tr2 SOM clustering label value L of Q fault signals q By comparing each l p With L q The predictive performance of the SOM model is determined by the degree of matching of the values ​​and the true label values; the GOA-SOM diagnostic model is applied to the test set S. te This leads to the final overall diagnostic results.

Description

Technical Field

[0001] This document relates to the field of tiltmeter technology, and in particular to a method and apparatus for diagnosing faults in a VP tiltmeter. Background Technology

[0002] The VP-type tiltmeter is a novel broadband vertical pendulum topographic deformation observation device independently developed. It has been installed and applied at several seismic stations (such as the Yixian Seismic Station in Hebei, the Jixian Seismic Station in Tianjin, and the Wuzhishan Deformation Station in Hainan) and some monitoring sites in the surveying and mapping industry, achieving certain results. However, as an early-developed electronic sensing device, it suffers from drawbacks such as insufficient experience in fault analysis, numerous fault and anomaly records, and extreme sensitivity to the observation environment. Furthermore, the current fault analysis and diagnosis of the VP-type tiltmeter mainly relies on manual judgment and traditional methods following established maintenance procedures. This often leads to ambiguous judgments, affecting the efficiency of instrument fault diagnosis and maintenance.

[0003] Severe tiltmeter malfunctions and signal anomalies can easily interfere with tilted solid tidal deformation monitoring data. If the type of fault signal is not effectively identified and addressed promptly, it may contaminate precursor observation records, leading to significant problems such as delayed earthquake early warning and inaccurate earthquake prediction. Therefore, rapid and accurate diagnosis of tiltmeter malfunctions at seismic stations is a prerequisite and is essential. Currently, the tiltmeters deployed at seismic stations are mainly of the VS, DSQ, and SSQ types. Although they differ slightly in observation principles, manufacturing dates, and instrument performance indicators, the operation, maintenance, and troubleshooting of these traditional types of tiltmeters have relatively mature supporting solutions. They can rely on experienced technical personnel and the technical support of instrument manufacturers to conduct reasonable fault diagnosis and troubleshooting. However, other newer deformation instruments (such as the VP type) have significant shortcomings in accurate fault diagnosis, lacking extensive instrument management experience and advanced fault diagnosis methods.

[0004] In summary, the shortcomings of the existing technology are as follows:

[0005] 1. Traditional VP-type tiltmeter fault diagnosis methods rely too much on human experience and manual operation to determine the type of instrument fault and the location of the lesion. This is highly subjective and prone to misjudgment and omission. In addition, manual repair takes too long and the process is too cumbersome.

[0006] 2. Some seismic station personnel even need to rely on previous VS-type tiltmeter failure analyses to infer the cause of VP tiltmeter failures, lacking a targeted methodology to effectively diagnose VP tiltmeter failures;

[0007] 3. Fault intelligent diagnosis technologies or traditional neural network models in other fields have certain defects and need to be improved in order to achieve more stable and accurate diagnostic results. Summary of the Invention

[0008] The purpose of this invention is to provide a method and apparatus for diagnosing faults in a VP tilt meter, thereby solving the aforementioned problems in the prior art.

[0009] This invention provides a method for diagnosing faults in a VP tiltmeter, comprising:

[0010] Preprocessing is performed on raw fault data from multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i The fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model;

[0011] The input matrix S0 is divided into training sets S according to a fixed ratio and random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into several parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the optimization of the GOA algorithm.

[0012] With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q ;

[0013] By comparing each l p With L qThe degree of matching of values ​​and the true label values ​​determine the prediction effect of the SOM model; and based on the prediction effect, it is determined whether the GOA iteration stopping condition is met and the iteration is stopped. The element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

[0014] The GOA-SOM diagnostic model was applied to the test set S. te This leads to the final overall diagnostic results.

[0015] This invention provides a fault diagnosis device for a VP tiltmeter, comprising:

[0016] The first processing module is used to preprocess the raw fault data of multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i The fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model;

[0017] The second processing module is used to divide the input matrix S0 into training sets S according to a fixed ratio and by random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into several parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the optimization of the GOA algorithm.

[0018] The third processing module is used to process the training set S. tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q ;

[0019] The fourth processing module is used to compare each l p With L qThe degree of matching of values ​​and the true label values ​​determine the prediction effect of the SOM model; and based on the prediction effect, it is determined whether the GOA iteration stopping condition is met and the iteration is stopped. The element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

[0020] The diagnostic module is used to apply the GOA-SOM diagnostic model to the test set S. te This leads to the final overall diagnostic results.

[0021] Compared with existing technologies, the VP tiltmeter fault diagnosis method proposed in this invention has high intelligence and automatic identification capabilities, effectively avoiding the uncertainty and error rate caused by manual diagnosis. It only requires classifying, labeling, and preprocessing historical fault data before diagnosis, and then the improved machine learning model can be used for fault diagnosis. Experimental results show that the method described in this invention has a high fault diagnosis rate and reliability, and the improved method is more accurate than traditional machine learning models, meeting the requirements of practicality and innovation. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart of the VP tilt meter fault diagnosis method according to an embodiment of the present invention;

[0024] Figure 2 This is a detailed flowchart of the VP tilt meter fault diagnosis method according to an embodiment of the present invention;

[0025] Figure 3 This is a schematic diagram of a typical VP tiltmeter fault signal according to an embodiment of the present invention;

[0026] Figure 4 This is a schematic diagram of VP tilt meter fault diagnosis according to an embodiment of the present invention. Detailed Implementation

[0027] The purpose of this invention is to provide a fault diagnosis method for VP-type tiltmeters based on artificial neural networks and GOA, changing the traditional method of relying mainly on manual inspection for VP tiltmeter fault diagnosis. Firstly, the concepts of empirical mode decomposition and distribution entropy are introduced to extract multi-scale features of instrument fault signals. Secondly, the key network parameters of the self-organizing map neural network are improved based on the locust optimization algorithm, and the improved model is used for intelligent fault diagnosis. This results in a newly designed diagnostic model with robustness and high diagnostic accuracy, which has certain application value in improving the efficiency of rapid fault diagnosis at monitoring stations.

[0028] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0029] Method Implementation Examples

[0030] According to an embodiment of the present invention, a method for diagnosing faults in a VP tiltmeter is provided. Figure 1 This is a flowchart of the VP tiltmeter fault diagnosis method according to an embodiment of the present invention, as follows: Figure 1 As shown, the VP tiltmeter fault diagnosis method according to an embodiment of the present invention specifically includes:

[0031] Step 101: Preprocess the raw fault data of multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i The fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model;

[0032] Step 102: Divide the input matrix S0 into training sets S according to a fixed ratio and by random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into several parts according to a fixed ratio during the optimization process. tr1 With test set S tr2The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the optimization of the GOA algorithm.

[0033] Step 103, using the training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q ;

[0034] Step 104, by comparing each l p With L q The prediction performance of the SOM model is determined by the degree of matching of the values ​​and the true label values ​​(and by designing a novel fitness function for the GOA algorithm); and the iteration is stopped based on the prediction performance to determine whether the GOA iteration stopping condition is met. The original parameter values ​​in the SOM model are replaced by the element values ​​of the obtained optimal solution to obtain a novel and reliable GOA-SOM diagnostic model.

[0035] Step 105: Apply the GOA-SOM diagnostic model to the test set S te This leads to the final overall diagnostic results.

[0036] The technical solutions of the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] As can be seen from the above description:

[0038] 1. This invention applies machine learning methods and intelligent diagnostic concepts to VP tiltmeter fault diagnosis;

[0039] 2. Optimize the network parameters of the SOM neural network (dimensionality of the first competitive layer, dimension of the second competitive layer, step size in the classification stage, and neighborhood distance in the tuning stage) using the Locust Optimization Algorithm (GOA) to construct a new model to improve its diagnostic performance.

[0040] 3. A novel feature criterion specifically designed for VP tiltmeter fault signals was developed. This involves performing Complete Ensemble Empirical Mode Decomposition (CEEMD) on the tiltmeter fault signals, calculating the distribution entropy of each of the multiple IMF components obtained from the decomposition, and obtaining the multi-scale distribution entropy of each IMF component. By traversing multiple fault signals, a complete set of new VP tiltmeter fault signal feature criterion matrices can be constructed, which can then be used as the input and validation dataset for the diagnostic model.

[0041] 4. A novel optimization algorithm fitness function was designed specifically for SOM neural networks. It fully utilizes the clustering and outlier concepts of label values ​​to construct a new function that can amplify the degree of failure in fault diagnosis. The smaller the value, the more accurate the model.

[0042] like Figure 2 As shown, the fault diagnosis process of the VP-type vertical pendulum tilter based on the CEEMD multi-scale distribution entropy and GOA-SOM neural network model of the present invention is as follows:

[0043] Step 1 involves performing data preprocessing operations such as distortion point localization, uniform length truncation, and data normalization on N raw fault data from VP-type tiltmeters (including three types of data: power failure, data acquisition failure, and environmental interference) to obtain the input vector X for fault feature analysis. i (i = 1, 2, ..., N), and its length is N;

[0044] Step 2 uses the Complete Ensemble Empirical Mode Decomposition (CEEMD) technique to decompose the i-th fault signal X. i Decomposed into 6 intrinsic mode functions (IMFs) i,j (j = 1, 2, ..., 6) and one residual signal R i And can be linearly reconstructed (restored) to X. i ,Right now

[0045] Step 3 calculates the distribution entropy of each IMF component to obtain the CEEMD multi-scale distribution entropy vector of the signal, and repeats steps 1-2 to calculate each fault signal X in turn. i The CEEMD multiscale distribution entropy value is used to construct the input matrix S0 of this model.

[0046] Step 4: Key Model Parameter Settings: Divide the input matrix S0 into training sets S0 by random sampling in a fixed ratio of 8:2 (or other ratios). tr and test set S te And the training set Str The training set S is randomly divided into two parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four key network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—were used as unknown variables for the optimization of the GOA algorithm.

[0047] Step 5: Execute the GOA algorithm: using the training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p (p = 1, 2, ..., P) (Label values ​​are all positive integers, and each fault signal of the same fault type generally has multiple different label values), such as Figure 3 As shown, the label values ​​belonging to "power failure" are recorded as "label". 1,i (i = 1, 2, ..., lb1), and denote the label values ​​that belong to "data acquisition failure" as label. 2,i (i = 1, 2, ..., lb2), and denote the label values ​​belonging to "environmental interference" as label. 3,i (i = 1, 2, ..., lb3), where lb1, lb2, and lb3 represent the number of labels for each fault type, and lb1 + lb2 + lb3 = P; and the test set S is applied. tr2 To perform the prediction (testing) process of the SOM model, the test set S can be obtained. tr2 SOM clustering label value L of Q fault signals q (q = 1, 2, ..., Q).

[0048] Step 6 involves comparing each l p With L q The degree of matching between the values ​​and the true label values ​​determines the prediction performance of SOM, i.e., L q Value and l p Values ​​are matched one-to-one; if the label values ​​are the same, and if l p With L q If the actual attributes of the fault signals they represent are the same, then the diagnosis is considered correct.

[0049] Otherwise, a diagnostic error is detected, and the label value L for that diagnostic failure is recorded. q Given the actual fault type k (k = 1, 2, or 3, where 1 represents "power failure", 2 represents "data acquisition failure", and 3 represents "environmental interference"), design the j-th diagnostic error label L. q Actual outlier coefficient (J represents the total number of fault diagnosis errors in this round, w) j(Initial values ​​are all 0); finally, by iterating through the diagnostic results of all Q fault signals, the overall accuracy of this diagnosis can be obtained as acc = 1 - J / Q, the number of diagnostic errors J, and w. j The value is calculated, and the fitness function f of GOA is calculated according to formula (1). The smaller the value of the f function, the more accurate the model prediction. If it is 0, it means that the accuracy of the diagnostic model is 100%.

[0050]

[0051] Step 7: If the GOA iteration stopping condition is met (Y value less than 0.001, or iterations completed 500 times), the loop is exited; otherwise, the above steps continue until the optimization process is forcibly terminated. Finally, the element values ​​of the obtained optimal solution replace the original parameter values ​​in the SOM model, resulting in a novel and reliable GOA-SOM diagnostic model. Due to the random sampling of the training and test sets, the datasets are different for each experiment, and the key parameter values ​​of the GOA-SOM model are not entirely consistent. Therefore, the model is dynamic and can effectively adapt to data changes.

[0052] Step 8: Apply the obtained GOA-SOM diagnostic model to the test set S. te This leads to the final overall diagnostic results.

[0053] Compared with existing technologies, the VP tiltmeter fault diagnosis method proposed in this invention has high intelligence and automatic identification capabilities, effectively avoiding the uncertainty and error rate caused by manual diagnosis. It only requires classifying, labeling, and preprocessing historical fault data before diagnosis, and then the improved machine learning model can be used for fault diagnosis. Experimental results show that the method described in this invention has a high fault diagnosis rate and reliability, and the improved method is more accurate than traditional machine learning models, meeting the requirements of practicality and innovation.

[0054] Device Examples

[0055] According to an embodiment of the present invention, a fault diagnosis device for a VP tiltmeter is provided. Figure 4 This is a schematic diagram of the VP tiltmeter fault diagnosis device according to an embodiment of the present invention, as shown below. Figure 4 As shown, the VP tilt meter fault diagnosis device according to an embodiment of the present invention specifically includes:

[0056] The first processing module 10 is used to preprocess the raw fault data of multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. iThe fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model;

[0057] The second processing module 12 is used to divide the input matrix S0 into training sets S0 according to a fixed ratio and by random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into two parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the GOA algorithm optimization. The fixed ratio is 8:2, but other ratios are also possible.

[0058] The third processing module 14 is used to process the training set S. tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q ;

[0059] The fourth processing module 16 is used to compare each l p With L q The degree of matching of values ​​and the true label values ​​determine the prediction effect of the SOM model; and based on the prediction effect, it is determined whether the GOA iteration stopping condition is met and the iteration is stopped. The element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

[0060] Diagnostic module 18 is used to apply the GOA-SOM diagnostic model to the test set S. te This leads to the final overall diagnostic results.

[0061] The first processing module 10 is specifically used for:

[0062] Step 1: Perform data preprocessing operations on N raw fault data of VP-type tiltmeters, including distortion point location, uniform length truncation, and data normalization, to obtain the input vector X for fault feature analysis. ii = 1, 2, ..., N, and its length is N; the original fault data specifically includes three types of data: power failure, data acquisition failure, and environmental interference;

[0063] Step 2: Use Complementary Set Empirical Mode Decomposition (CEEMD) to analyze the i-th fault signal X. i Decomposed into 6 intrinsic mode functions (IMFs) i,j j = 1, 2, ..., 6 and a residual signal R i And linearly reconstructed into X i ,Right now

[0064] Step 3: Calculate the distribution entropy of each IMF component to obtain the CEEMD multi-scale distribution entropy vector of the fault signal, and repeat steps 1 and 2 to calculate each fault signal X in turn. i The CEEMD multiscale distribution entropy value is used to construct the input matrix S0 of the model.

[0065] The third processing module 14 is specifically used for:

[0066] With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p p = 1, 2, ..., P, where the label values ​​are all positive integers, and each fault signal of the same fault type generally has multiple different label values. The label value belonging to "power supply fault" is denoted as label. 1,i (i = 1, 2, ..., lb1), and denote the label values ​​that belong to "data acquisition failure" as label. 2,i (i = 1, 2, ..., lb2), and denote the label values ​​belonging to "environmental interference" as label. 3,i (i = 1, 2, ..., lb3), where lb1, lb2, and lb3 represent the number of labels for each fault type, and lb1 + lb2 + lb3 = P; and the test set S is applied. tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q , q=1,2,…,Q.

[0067] The fourth processing module 16 is specifically used for:

[0068] By comparing each l p With L q The degree of matching between the values ​​and the true label values ​​determines the prediction performance of SOM, i.e., L q Value and l pValues ​​are matched one-to-one; if the label values ​​are the same, and if l p With L q If the actual attributes of the fault signals they represent are the same, then the diagnosis is considered correct; otherwise, the diagnosis is incorrect, and the label value L for the failed diagnosis is determined accordingly. q And its actual fault type k, where k = 1, 2, or 3, where 1 represents "power supply fault", 2 represents "data acquisition fault", and 3 represents "environmental interference". Design the j-th diagnostic error label L. q Actual outlier coefficient Where J represents the total number of fault diagnosis errors in this round, w j The initial values ​​are all 0; finally, by iterating through the diagnostic results of all Q fault signals, the overall accuracy of this diagnosis, acc = 1 - J / Q, the number of diagnostic errors J, and w can be obtained. j The value is calculated, and the fitness function f of GOA is calculated according to Formula 1. The smaller the value of the f function, the more accurate the model prediction. If it is 0, it means that the accuracy of the diagnostic model is 100%.

[0069]

[0070] If the GOA iteration stopping condition is met, the loop is exited; otherwise, the above steps are continued until the optimization process is forcibly terminated. Finally, the element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

[0071] The embodiments of the present invention are device embodiments corresponding to the above method embodiments. The specific operation of each module can be understood with reference to the description of the method embodiments, and will not be repeated here.

[0072] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing faults in a VP tiltmeter, characterized in that, include: Preprocessing is performed on raw fault data from multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i The fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model; The input matrix S0 is divided into training sets S according to a fixed ratio and by random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into two parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the optimization of the GOA algorithm. With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals p ; By comparing each l p With L p The degree of matching of values ​​and the true label values ​​determine the prediction effect of the SOM model; and based on the prediction effect, it is determined whether the GOA iteration stopping condition is met and the iteration is stopped. The element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model. The GOA-SOM diagnostic model was applied to the test set S. te This leads to the final overall diagnostic results.

2. The method according to claim 1, characterized in that, Preprocessing is performed on raw fault data from multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i It is decomposed into 6 intrinsic mode functions and 1 residual signal, and linearly reconstructed into X. i Calculate the distribution entropy of each intrinsic mode function to obtain the CEEMD multi-scale distribution entropy vector of the fault signal, and then calculate the distribution entropy of each fault signal X in sequence. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model, which specifically includes: Step 1: Perform data preprocessing operations on N raw fault data of VP-type tiltmeters, including distortion point location, uniform length truncation, and data normalization, to obtain the input vector X for fault feature analysis. i i = 1, 2, ..., N, and its length is N; the original fault data specifically includes three types of data: power failure, data acquisition failure, and environmental interference; Step 2: Use Complementary Set Empirical Mode Decomposition (CEEMD) to analyze the i-th fault signal X. i Decomposed into 6 intrinsic mode functions (IMFs) i,j j = 1, 2, ..., 6 and a residual signal R i And linearly reconstructed into X i ,Right now Step 3: Calculate the distribution entropy of each IMF component to obtain the CEEMD multi-scale distribution entropy vector of the fault signal, and repeat steps 1 and 2 to calculate each fault signal X in turn. i The CEEMD multiscale distribution entropy value is used to construct the input matrix S0 of the model.

3. The method according to claim 1, characterized in that, With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q Specifically, it includes: With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p p = 1, 2, ..., P, where the label values ​​are all positive integers, and each fault signal of the same fault type generally has multiple different label values. The label value belonging to "power supply fault" is denoted as label. 1,i (i = 1, 2, ..., lb1), and denote the label values ​​that belong to "data acquisition failure" as label. 2,i (i = 1, 2, ..., lb2), and denote the label values ​​belonging to "environmental interference" as label. 3,i (i = 1, 2, ..., lb3), where lb1, lb2, and lb3 represent the number of labels for each fault type, and lb1 + lb2 + lb3 = P; and the test set S is applied. tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q , q=1,2,…,Q.

4. The method according to claim 1, characterized in that, By comparing each l p With L q The predictive performance of the SOM model is determined by the degree of matching of values ​​and the true label values; and the iteration is stopped based on the predictive performance to determine whether the GOA iteration stopping condition is met. The original parameter values ​​in the SOM model are replaced with the element values ​​of the obtained optimal solution to obtain a new and reliable GOA-SOM diagnostic model, which specifically includes: By comparing each l p With L q The degree of matching between the values ​​and the true label values ​​determines the prediction performance of SOM, i.e., L q Value and l p Values ​​are matched one-to-one; if the label values ​​are the same, and if l p With L q If the actual attributes of the fault signals they represent are the same, then the diagnosis is considered correct; otherwise, the diagnosis is incorrect, and the label value L for the failed diagnosis is determined accordingly. q And its actual fault type k, where k = 1, 2, or 3, where 1 represents "power supply fault", 2 represents "data acquisition fault", and 3 represents "environmental interference". Design the j-th diagnostic error label L. q Actual outlier coefficient Where J represents the total number of fault diagnosis errors in this round, w j The initial values ​​are all 0; finally, by iterating through the diagnostic results of all Q fault signals, the overall accuracy of this diagnosis, acc = 1 - JQ, the number of diagnostic errors J, and w can be obtained. j The value is calculated, and the fitness function f of GOA is calculated according to Formula 1. The smaller the value of the f function, the more accurate the model prediction. If it is 0, it means that the accuracy of the diagnostic model is 100%. If the GOA iteration stopping condition is met, the loop is exited; otherwise, the above steps are continued until the optimization process is forcibly terminated. Finally, the element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

5. A fault diagnosis device for a VP tiltmeter, characterized in that, include: The first processing module is used to preprocess the raw fault data of multiple VP-type tiltmeters to obtain the input vector X for fault feature analysis. i The complementary set empirical mode decomposition technique is used to analyze the i-th fault signal X. i The fault signal is decomposed into six intrinsic mode functions (EMFs) and one residual signal, and then subjected to CEEMD decomposition to obtain several EEMFs. The distribution entropy of each EEMF is calculated to obtain the CEEMD multi-scale distribution entropy vector of the fault signal. Then, the distribution entropy vectors of each fault signal X are calculated sequentially. i The CEEMD multi-scale distribution entropy value is used to construct the input matrix S0 of this model; The second processing module is used to divide the input matrix S0 into training sets S according to a fixed ratio and by random sampling. tr and test set S te And the training set S tr The training set S is randomly divided into two parts according to a fixed ratio during the optimization process. tr1 With test set S tr2 The four important network parameters of the SOM neural network model—the dimension of the first competitive layer, the dimension of the second competitive layer, the step size of the classification stage, and the neighborhood distance of the tuning stage—are used as unknown variables for the optimization of the GOA algorithm. The third processing module is used to process the training set S. tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p and through the application test set S tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q ; The fourth processing module is used to compare each l p With L q The degree of matching of values ​​and the true label values ​​determine the prediction effect of the SOM model; and based on the prediction effect, it is determined whether the GOA iteration stopping condition is met and the iteration is stopped. The element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model. The diagnostic module is used to apply the GOA-SOM diagnostic model to the test set S. te This leads to the final overall diagnostic results.

6. The apparatus according to claim 5, characterized in that, The first processing module is specifically used for: Step 1: Perform data preprocessing operations on N raw fault data of VP-type tiltmeters, including distortion point location, uniform length truncation, and data normalization, to obtain the input vector X for fault feature analysis. i i = 1, 2, ..., N, and its length is N; the original fault data specifically includes three types of data: power failure, data acquisition failure, and environmental interference; Step 2: Use Complementary Set Empirical Mode Decomposition (CEEMD) to analyze the i-th fault signal X. i Decomposed into 6 intrinsic mode functions (IMFs) i,j j = 1, 2, ..., 6 and a residual signal R i And linearly reconstructed into X i ,Right now Step 3: Calculate the distribution entropy of each IMF component to obtain the CEEMD multi-scale distribution entropy vector of the fault signal, and repeat steps 1 and 2 to calculate each fault signal X in turn. i The CEEMD multiscale distribution entropy value is used to construct the input matrix S0 of the model.

7. The apparatus according to claim 5, characterized in that, The third processing module is specifically used for: With training set S tr1 The SOM model is trained on the object to obtain the training set S. tr1 SOM clustering label values ​​l of P fault signals p p = 1, 2, ..., P, where the label values ​​are all positive integers, and each fault signal of the same fault type generally has multiple different label values. The label value belonging to "power supply fault" is denoted as label. 1,i (i = 1, 2, ..., lb1), and denote the label values ​​that belong to "data acquisition failure" as label. 2,i (i = 1, 2, ..., lb2), and denote the label values ​​belonging to "environmental interference" as label. 3,i (i = 1, 2, ..., lb3), where lb1, lb2, and lb3 represent the number of labels for each fault type, and lb1 + lb2 + lb3 = P; and the test set S is applied. tr2 To perform the prediction process of the SOM model, the test set S is obtained. tr2 SOM clustering label value L of Q fault signals q , q=1,2,…,Q.

8. The apparatus according to claim 5, characterized in that, The fourth processing module is specifically used for: By comparing each l p With L q The degree of matching between the values ​​and the true label values ​​determines the prediction performance of SOM, i.e., L q Value and l p Values ​​are matched one-to-one; if the label values ​​are the same, and if l p With L q If the actual attributes of the fault signals they represent are the same, then the diagnosis is considered correct. Otherwise, a diagnostic error is detected, and the label value L for that diagnostic failure is recorded. q And its actual fault type k, where k = 1, 2, or 3, where 1 represents "power supply fault", 2 represents "data acquisition fault", and 3 represents "environmental interference". Design the j-th diagnostic error label L. q Actual outlier coefficient Where J represents the total number of fault diagnosis errors in this round, w j The initial values ​​are all 0; finally, by iterating through the diagnostic results of all Q fault signals, the overall accuracy of this diagnosis, acc = 1 - JQ, the number of diagnostic errors J, and w can be obtained. j The value is calculated, and the fitness function f of GOA is calculated according to Formula 1. The smaller the value of the f function, the more accurate the model prediction. If it is 0, it means that the accuracy of the diagnostic model is 100%. If the GOA iteration stopping condition is met, the loop is exited; otherwise, the above steps are continued until the optimization process is forcibly terminated. Finally, the element values ​​of the obtained optimal solution are used to replace the original parameter values ​​in the SOM model to obtain a new and reliable GOA-SOM diagnostic model.

Images