High-voltage switch cabinet fault diagnosis intelligent decision-making method based on LPWAN multi-source information fusion

Abstract

The invention discloses a high-voltage switch cabinet fault diagnosis intelligent decision-making method based on LPWAN multi-source information fusion. The method comprises the following steps: acquiring operation state data of a high-voltage switch cabinet; building an ad hoc network according to the LPWAN technology, and sending the operation state data collected by the sensing layer to a decision layer through the ad hoc network; performing data standardization processing on the operation state data through a decision-making layer, and performing feature fusion on the standardized data; combining a principal component analysis (PCA) with a probabilistic neural network (PNN) to construct a multi-source information fusion fault diagnosis model PCA-PNN; and performing analysis and judgment on the data after feature fusion through PCA-PNN to obtain an operation state of the high-voltage switch cabinet, judging whether a fault occurs or not and a fault type according to the operation state of the high-voltage switch cabinet, and generating a corresponding decision according to the fault type. According to the method, fault feature extraction is carried out in combination with a principal component analysis method, the PCA-PNN model is constructed, the fault diagnosis period is shortened, and the working efficiency is improved; and in combination with the LPWAN, the effects of low power consumption and wide coverage are achieved.

Description

Technical Field

[0001] The present invention belongs to the field of high-voltage switch cabinet fault diagnosis, and specifically relates to a high-voltage switch cabinet fault diagnosis intelligent decision-making method based on LPWAN multi-source information fusion. Background Art

[0002] High-voltage switchgear plays a key role in the power transmission and distribution system, responsible for the opening and closing of power lines and monitoring of operating power data. With the rapid progress of power systems towards intelligence, the demand for stability and safety of high-voltage switchgear is increasing, and the existing single-function switchgear detection equipment can no longer meet these needs. In previous studies, a design of an online comprehensive monitoring device for high-voltage switchgear based on the Internet of Things has been proposed. This device integrates multiple detection units in the cabinet to achieve real-time monitoring of the high-voltage switchgear, but lacks data processing and analysis functions.

[0003] Probabilistic neural network (PNN) is a parallel classification algorithm based on the Bayesian minimum risk criterion. It has excellent nonlinear classification capabilities and has been widely used in electrical equipment fault diagnosis. Previous studies have proposed a PNN-based permanent magnet synchronous linear motor partial demagnetization fault diagnosis method and a PNN-based high-voltage circuit breaker fault diagnosis method. However, due to the large number of electrical components and fault types inside the high-voltage hall switch cabinet, there may be correlation or redundancy between the feature quantities, and the computational complexity of the PNN network is large, resulting in a long fault diagnosis cycle. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion, combine the principal component analysis method PCA to extract fault features, and construct a PCA-PNN model, which can effectively shorten the fault diagnosis cycle and improve work efficiency; and combine with low-power wide area network LPWAN to achieve the effects of low power consumption, wide coverage and flexible deployment.

[0005] In order to solve the above technical problems, the technical solution of the present invention is: a high-voltage switch cabinet fault diagnosis intelligent decision-making method based on LPWAN multi-source information fusion, which is applied to a high-voltage switch cabinet fault diagnosis intelligent decision-making system based on LPWAN multi-source information fusion, the system at least includes a perception layer for collecting high-voltage switch cabinet operation status data, a transmission layer for data uploading, and a decision layer for fault judgment and decision output according to the operation status data; the method includes the following steps:

[0006] S1. Data collection: collect the operating status data of the high-voltage switchgear through the sensing layer; the operating status data at least includes closing current, opening current, bus current and voltage, and contact temperature;

[0007] S2, data upload: build a self-organizing network based on LPWAN technology, and use the self-organizing network as the communication network of the transmission layer. The transmission layer sends the operating status data collected by the perception layer to the decision layer through the self-organizing network;

[0008] S3, data processing and model building: The decision layer performs data standardization on the operating status data and performs feature fusion on the standardized data; the principal component analysis method PCA and the probabilistic neural network PNN are combined to construct a multi-source information fusion fault diagnosis model PCA-PNN, and PCA-PNN is used as the basis for fault judgment and decision output of the decision layer;

[0009] S4. Fault judgment and decision output: The multi-source information fusion fault diagnosis model PCA-PNN is used to analyze and judge the data after feature fusion to obtain the operating status of the high-voltage switchgear. According to the operating status of the high-voltage switchgear, it is judged whether a fault has occurred and the type of fault, and the corresponding decision is generated according to the fault type.

[0010] S1 is specifically:

[0011] S11. Set the frequency of collecting running status data according to actual needs;

[0012] S12, preprocessing the collected operating status data, where the preprocessing at least includes noise removal, data calibration and data alignment;

[0013] S13, according to the properties of the high-voltage switchgear and the fault type, the features related to the fault judgment are extracted through Fourier transform;

[0014] S14. After cleaning, denoising and normalizing the features, they are used as input of the fault feature data set.

[0015] S2 is specifically:

[0016] S21. Build a self-organizing network based on LPWAN technology;

[0017] S22, connecting the data collection equipment of the perception layer to the ad hoc network;

[0018] S23, uploading the operation status data through the ad hoc network;

[0019] S24. The decision-making layer receives the operation status data.

[0020] S3 is specifically:

[0021] S31, performing data standardization processing on the operation status data, the method of data standardization processing is:

[0022]

[0023] Among them, R k represents the running status data matrix, r ij is the running status data of the i-th row and j-th column in the running status data matrix;

[0024] The element r ij (1≤i≤m,1≤j≤n) is standardized as follows:

[0025]

[0026] in, is the result of standardization, r j is the running status data matrix R k The mean of the jth column, s j is the standard deviation of the jth column;

[0027] Depend on Construct a new feature matrix Right now:

[0028]

[0029] Among them, m and n represent the number of training samples and the dimension of features respectively;

[0030] S32, through the principal component analysis PCA feature matrix Perform feature extraction to reduce the dimension of the operating status data;

[0031] S33, constructing a multi-source information fusion fault diagnosis model PCA-PNN: adopting a probabilistic neural network PNN network structure, the network structure includes an input layer, a pattern layer, a summation layer and an output layer; wherein,

[0032] The input layer is used to transform the fault feature dataset Y = [y1, y2, ..., y k ] After assigning weights, they are passed to all neurons; neurons are the basic units of a neural network, which are used to receive input signals, weight the inputs, and then generate output signals through activation functions. The total number of neurons is the length of the feature vector, and the feature vector represents the data input into the model and used to reflect the fault characteristics;

[0033] The pattern layer is used to calculate the matching relationship between the feature vector and each fault mode in the training sample. The fault mode refers to the different types of faults collected in the training sample. Each fault mode corresponds to a neuron in the pattern layer.

[0034] The summation layer is used to calculate the probability of the feature vector corresponding to each fault mode and accumulate the probability of it belonging to a certain type of fault;

[0035] The output layer is composed of a threshold discriminator, which is used to compare the probability density functions under various fault modes, set the neuron output with the largest function value to 1, and the outputs of other neurons to 0, and classify the fault type by comparing the probability values.

[0036] S34, input the feature extraction result to the probabilistic neural network PNN.

[0037] S32 is specifically:

[0038] For the feature matrix The correlation between all elements in is calculated, and the corresponding correlation coefficient matrix is ​​expressed as:

[0039]

[0040]

[0041] Among them, c ij For the matrix Any two variables and The correlation coefficient, x l For the matrix The mean of the i-th row, x J for The mean of the jth column;

[0042] For the correlation coefficient matrix C mn Perform eigenvalue decomposition, calculate the eigenvalue j according to the characteristic equation, and sort them by size. The corresponding eigenvalue is p j , calculate the main element t j And the contribution rate of each main element is expressed as:

[0043]

[0044]

[0045] Combined with practical applications, the principal components corresponding to the eigenvalues ​​with contribution rates greater than 85% are selected to construct a new sample feature data set, which is then input into the PNN network for fault diagnosis.

[0046] The output of each mode unit in S33 is:

[0047]

[0048] Among them, W k is the connection weight between the input layer and the pattern layer; δ is the smoothing factor.

[0049] The probability of belonging to a certain type of fault in S33 is accumulated in the following way:

[0050]

[0051] Among them, f A is the failure probability, W ak is the kth training vector of the fault mode; ρ is the data dimension of the sample space.

[0052] The probability density function of each failure mode is expressed as:

[0053] M = arg max(f A )

[0054] Where M is the output of the fault type.

[0055] A high-voltage switch cabinet fault diagnosis intelligent decision-making system based on LPWAN multi-source information fusion is also provided, including:

[0056] The perception layer is used to collect the operating status data of the high-voltage switchgear;

[0057] Transport layer, used to upload operation status data;

[0058] The decision layer is used to make fault judgment and decision output according to the operation status data, wherein the decision layer stores the intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion as described in any of the above items.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] The present invention combines the principal component analysis method PCA to extract fault features and constructs a PCA-PNN model, which can effectively shorten the fault diagnosis cycle and improve work efficiency; and combines it with the low-power wide area network LPWAN to achieve the effects of low power consumption, wide coverage and flexible deployment. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a structural schematic diagram of an embodiment of the present invention;

[0062] Figure 2 Schematic diagram of the structure of the PNN model in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0064] See also Figure 1The technical solution of the present invention is: a framework of a high-voltage switchgear fault diagnosis intelligent decision-making system based on LPWAN multi-source information fusion, comprising the following steps:

[0065] S1. Data acquisition. The system collects the operating status data of the high-voltage switchgear. This may include various monitoring parameters such as current, voltage, temperature, etc.

[0066] S2, data upload. The system builds an ad hoc network through LPWAN technology to upload data. This step can realize communication between devices and real-time data transmission.

[0067] S3. Data processing and model building. A management system with B / S architecture is built for human-computer interaction and alarm analysis. The management system performs standardized preprocessing on the original monitoring data of the switch cabinet, and then uses PCA (principal component analysis) to fuse the features of the samples, uses PNN (probabilistic neural network) for comprehensive decision analysis, and establishes a PCA-PNN multi-source information fusion fault diagnosis model.

[0068] S4. Fault judgment and decision-making. The system judges the operating status of the switchgear according to the above model, and makes reasonable decisions to deal with the fault according to the diagnosis results, thereby improving the automation and intelligence level of emergency response of high-voltage switchgear and reducing the social and economic impact of switchgear accidents.

[0069] In S1, data collection specifically includes:

[0070] S11. Set the data collection frequency according to actual needs to obtain sufficient data.

[0071] S12. Preprocessing the collected raw data, including steps such as noise removal, data calibration and data alignment.

[0072] S13. According to the characteristics and fault types of the high-voltage switchgear, key features extracted using Fourier transform are as follows: equipment status, temperature, humidity, smoke concentration, etc.

[0073] S14. Perform operations such as cleaning, denoising, and normalization on the extracted key data features for subsequent processing.

[0074] In S2, data uploading is specifically as follows:

[0075] S21. Network construction. Use LPWAN technology to build a self-organizing network. This is achieved through the Internet of Things technology, which is characterized by low power consumption and long-distance communication, and is suitable for the construction of sensor networks.

[0076] S22, device access. Connect the data collection device to the LPWAN network. This may involve configuration and installation of the device to ensure that the device can connect to the network normally.

[0077] S23, data transmission. The switch cabinet operating status data is uploaded through the LPWAN network. This step needs to ensure the stability and security of data transmission to avoid data loss or tampering.

[0078] S24, data reception. After the data is uploaded, the management system receives and processes the data for further analysis and processing. It may be necessary to set specific interfaces and protocols to receive and process the data.

[0079] In S3, data processing and model building are specifically as follows:

[0080] S31, data standardization processing. The management system first performs standardization preprocessing on the collected original monitoring data of the switch cabinet. Suppose the original sample data matrix is ​​R k , as shown below:

[0081]

[0082] Each element r ij (1≤i≤m,1≤j≤n) is standardized as follows:

[0083]

[0084] In the formula, r j is the matrix R k The mean of the jth column, s j is the standard deviation of the jth column. Form a new matrix Right now:

[0085]

[0086] Where: m and n represent the number of samples and feature dimensions respectively.

[0087] S32, PCA feature fusion. Principal component analysis (PCA) can be used to extract the main elements of the sample set, reduce the feature dimension, and improve classification efficiency. The standardized matrix There is still a certain amount of redundancy, and it is necessary to extract features and reduce the data dimension.

[0088] For the feature matrix By calculating the correlation between all elements in , we can get the corresponding correlation coefficient matrix:

[0089]

[0090]

[0091] Where: cij For the matrix Any two variables and The correlation coefficient, x ι For the matrix The mean of the i-th row, x J for The mean of the jth column. The correlation coefficient matrix C mn Perform eigenvalue decomposition, calculate the eigenvalue j according to the characteristic equation, and sort them by size. The corresponding eigenvalue is p j , and finally calculate the main element t j And the contribution rate of each principal component. The formula is as follows:

[0092]

[0093]

[0094] S33, analysis model construction. Use PNN for comprehensive decision analysis. PNN is a neural network based on Bayesian theory, which can be used for classification and prediction. By training the PNN model, a fault diagnosis model for high-voltage switchgear can be obtained.

[0095] S43, PNN decision making. The structural model of PNN consists of input layer, pattern layer, summation layer and output layer, such as Figure 2 shown.

[0096] The input layer converts the sample feature data set Y = [y1, y2, ..., y k ] is assigned weights and passed to all pattern units. The number of neurons is the length of the input feature vector.

[0097] The pattern layer calculates the matching relationship between the input fault feature vector and each pattern in the training sample. The output of each pattern unit is:

[0098]

[0099] Where: W k is the connection weight between the input layer and the pattern layer; δ is the smoothing factor.

[0100] The summation layer calculates the probability of each fault mode in the sample and accumulates the probability of it belonging to a certain type of fault. That is:

[0101]

[0102] Where: W ak is the kth training vector of the fault mode; m is the number of fault mode training samples; ρ is the data dimension of the sample space; δ is the smoothing factor.

[0103] The output layer of the network is composed of threshold discriminators. By comparing the probability density functions under various fault modes, the output of the neuron with the largest function value is set to 1, and the outputs of other neurons are set to 0. That is:

[0104] M = argmax(f A )

[0105] Where: f A is the probability of failure; M is the fault type output. The classification output of switch cabinet faults is achieved by comparing the probability values.

[0106] In S4, fault judgment and decision making include the following steps:

[0107] S41, operating status judgment. The operating status of the high-voltage switchgear is judged through the PCA-PNN multi-source information fusion fault diagnosis model. This step may involve the operation of the model, the interpretation of the prediction results, etc.

[0108] S42, fault diagnosis. If the model determines that the switch cabinet is in an abnormal operating state, further diagnose the possible fault type. It may be necessary to deeply understand and identify various types of faults in order to make accurate diagnosis. The entire process of fault diagnosis is information preprocessing, PCA feature data fusion, and PNN network decision.

[0109] S43. Emergency response. Implement response strategies and conduct emergency response for high-voltage switchgear. This includes measures such as fault repair and equipment replacement to restore normal operation of the switchgear as soon as possible and reduce the social and economic impact of the accident.

[0110] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. An intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion, characterized in that: Applied to a high-voltage switchgear fault diagnosis intelligent decision-making system based on LPWAN multi-source information fusion, the system at least includes a perception layer for collecting high-voltage switchgear operating status data, a transmission layer for uploading data, and a decision-making layer for fault judgment and decision output based on the operating status data; the method includes the following steps: S1. Data collection: The operating status data of the high-voltage switchgear is collected through the sensing layer; the operating status data at least includes closing current, opening current, bus current and voltage, and contact temperature; S2, data upload: build a self-organizing network based on LPWAN technology, and use the self-organizing network as the communication network of the transmission layer. The transmission layer sends the operating status data collected by the perception layer to the decision layer through the self-organizing network; S3, data processing and model building: The decision layer performs data standardization on the operating status data and performs feature fusion on the standardized data; the principal component analysis method PCA and the probabilistic neural network PNN are combined to construct a multi-source information fusion fault diagnosis model PCA-PNN, and PCA-PNN is used as the basis for fault judgment and decision output of the decision layer; S4. Fault judgment and decision output: The multi-source information fusion fault diagnosis model PCA-PNN is used to analyze and judge the data after feature fusion to obtain the operating status of the high-voltage switchgear. According to the operating status of the high-voltage switchgear, it is judged whether a fault has occurred and the type of fault, and the corresponding decision is generated according to the fault type.

2. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 1 is characterized in that: S1 is specifically: S11. Set the frequency of collecting running status data according to actual needs; S12, preprocessing the collected operating status data, where the preprocessing at least includes noise removal, data calibration and data alignment; S13, according to the properties of the high-voltage switchgear and the fault type, the features related to the fault judgment are extracted through Fourier transform; S14. After cleaning, denoising and normalizing the features, they are used as input of the fault feature data set.

3. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 1 is characterized in that: S2 is specifically: S21. Build a self-organizing network based on LPWAN technology; S22, connecting the data collection equipment of the perception layer to the ad hoc network; S23, uploading the operation status data through the ad hoc network; S24. The decision-making layer receives the operation status data.

4. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 2 is characterized in that: S3 is specifically: S31, performing data standardization processing on the operation status data, the method of data standardization processing is: Among them, R k represents the running status data matrix, r ij is the running status data of the i-th row and j-th column in the running status data matrix; The element r ij (1≤i≤m,1≤j≤n) is standardized as follows: in, is the result of standardization, r j is the running status data matrix R k The mean of the jth column, s j is the standard deviation of the jth column; Depend on Construct a new feature matrix Right now: Among them, m and n represent the number of training samples and the dimension of features respectively; S32, through the principal component analysis PCA feature matrix Perform feature extraction to reduce the dimension of the operating status data; S33, constructing a multi-source information fusion fault diagnosis model PCA-PNN: adopting the network structure of the probabilistic neural network PNN, the network structure includes an input layer, a pattern layer, a summation layer and an output layer; wherein, The input layer is used to transform the fault feature dataset Y = [y1, y2, ..., y k ] is weighted and then passed to all neurons; neurons are the basic units of neural networks, which are used to receive input signals, weight the inputs and then generate output signals through activation functions; the total number of neurons is the length of the feature vector, and the feature vector represents the data input to the model and used to reflect the fault characteristics; The pattern layer is used to calculate the matching relationship between the feature vector and each fault mode in the training sample; the fault mode refers to the different types of faults collected in the training sample, and each fault mode corresponds to a neuron in the pattern layer; The summation layer is used to calculate the probability of each fault mode corresponding to the feature vector and accumulate the probability of it belonging to a certain type of fault; The output layer is composed of a threshold discriminator, which is used to compare the probability density functions under various fault modes, set the output of the neuron with the largest function value to 1, and the output of other neurons to 0, and classify and output the fault type by comparing the probability values; S34, input the feature extraction result to the probabilistic neural network PNN.

5. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 4 is characterized in that: S32 is specifically: For the feature matrix The correlation between all elements in is calculated, and the corresponding correlation coefficient matrix is ​​expressed as: Among them, c ij For the matrix Any two variables and The correlation coefficient, x i For the matrix The mean of the i-th row, x j for The mean of the jth column; For the correlation coefficient matrix C mn Perform eigenvalue decomposition, calculate the eigenvalue j according to the characteristic equation, and sort them by size. The corresponding eigenvalue is p j , calculate the main element t j And the contribution rate of each main element is expressed as: The principal components corresponding to the eigenvalues ​​with contribution rates greater than 85% are selected to construct a new sample feature data set, which is then input into the PNN network for fault diagnosis.

6. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 4 is characterized in that: The output of each mode unit in S33 is: Among them, W k is the connection weight between the input layer and the pattern layer; δ is the smoothing factor.

7. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 6 is characterized in that: The probability of belonging to a certain type of fault in S33 is accumulated in the following way: Among them, f A is the failure probability, W ak is the kth training vector of the fault mode; ρ is the data dimension of the sample space.

8. The intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion according to claim 7 is characterized in that: The probability density function of each failure mode is expressed as: M=arg max(f A ) Where M is the output of the fault type.

9. High-voltage switchgear fault diagnosis intelligent decision-making system based on LPWAN multi-source information fusion, characterized by: include: The perception layer is used to collect the operating status data of the high-voltage switchgear; Transport layer, used to upload operation status data; The decision layer is used to make fault judgment and decision output according to the operation status data, wherein the decision layer stores the intelligent decision-making method for high-voltage switchgear fault diagnosis based on LPWAN multi-source information fusion as described in any one of claims 1 to 8.

Images