High-voltage power distribution cabinet multi-dimensional data state evaluation method based on WOA-LSSVM model
Through the multi-dimensional data state evaluation method of high-voltage distribution cabinet based on WOA-LSSVM model, the problems of single data and low fault diagnosis accuracy in the existing technology are solved, and the rapid and accurate evaluation and fault detection of the status of high-voltage distribution cabinet are achieved. , ensuring the safe operation of the equipment.
Patent Information
- Application Number
- CN202411036920.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-31
- Publication Date
- 2024-12-06
AI Technical Summary
The fault judgment process data of the existing high-voltage distribution cabinet is single, the detection function is single, and the characteristics of the input data cannot be effectively extracted. The fault diagnosis accuracy is not high, and it cannot meet the monitoring and fault analysis requirements of the high-voltage distribution cabinet.
The multidimensional data state evaluation method of high-voltage distribution cabinet based on WOA-LSSVM model is adopted. Multidimensional data is collected by installing sensors, feature vectors are extracted using Fbank technology, and WOA-LSSVM model is constructed for training and verification, so as to achieve the depth of the state of high-voltage distribution cabinet. Mining and anomaly detection.
It realizes a rapid and accurate assessment of the status of high-voltage distribution cabinets, can promptly detect abnormal working conditions, provide theoretical basis for fault handling, and ensure safe operation of the equipment.
Smart Images

Figure CN119090289A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of high-voltage power distribution cabinets, and in particular to a state evaluation method of multi-dimensional data of a high-voltage power distribution cabinet based on a WOA-LSSVM model. Background Art
[0002] High-voltage distribution cabinet refers to a device that is assembled according to certain technical requirements, including busbars, switchgear, instruments, and transformers, and is used to receive, distribute, and control electrical energy. Common faults of high-voltage distribution cabinets include broken shells, oil and gas leakage, large sparks in bushing connections, causing flashover grounding, large gas leakage in pneumatic operating mechanisms, oil pressure drop in hydraulic operating mechanisms, hissing sounds from vacuum circuit breakers, spark discharges and overheating and smoking, and electromagnetic sounds in the secondary open circuit of current transformers. Therefore, it is of great significance to ensure the healthy operation of high-voltage distribution cabinets and to promptly discover and eliminate their abnormal operating conditions. At present, the fault judgment process of high-voltage distribution cabinets is to collect operating parameters and compare them with parameter thresholds to determine whether a fault has occurred. When the upper threshold ≤ real-time value ≤ lower threshold, it is judged to be in a normal state; otherwise, it is judged to be in a fault state. The existing technology has the following disadvantages: (1) The traditional high-voltage distribution cabinet cannot meet the requirements of monitoring and fault analysis of high-voltage distribution cabinets due to the single data collected, single detection function, and small amount of monitoring information; (2) In the existing technology, the existing high-voltage distribution cabinet cannot effectively extract the characteristics of the input data, and the fault diagnosis accuracy is not high. Summary of the invention
[0003] The technical problem to be solved by the present invention is to provide a state assessment method for multi-dimensional data of high-voltage distribution cabinets based on the WOA-LSSVM model, to perform fusion analysis and deep mining on real-time state monitoring data, to discover valuable knowledge from the perspective of analyzing the inherent laws of the data itself, to achieve rapid detection of abnormal states, to comprehensively, timely and accurately grasp the health status of key performance indicators, to provide a theoretical basis for fault handling, and to provide strong support for ensuring the safe operation of equipment.
[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: a state evaluation method of multidimensional data of high-voltage distribution cabinet based on WOA-LSSVM model, comprising the following steps: Step 1: According to the nameplate of the high-voltage distribution cabinet and manual experience, determine the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions; Step 2: According to the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions determined in step 1, a series of sensors are installed at corresponding positions of the high-voltage distribution cabinet to achieve the purpose of extracting various important parameter indicators; Step 3: Under normal operation of the high-voltage distribution cabinet, the data information collected by the voiceprint sensor and the vibration sensor installed in step 2 is processed using the Fbank technology to obtain the corresponding feature vector; Step 4: When the high-voltage distribution cabinet is in normal operation, the original data set collected by the computer monitoring system (voltage U, current I, temperature of each component, humidity of each component, and change range of characteristic gas concentration in the distribution cabinet during operation of the high-voltage distribution cabinet) is collected, and the original characteristic data set is cleaned by using the machine learning model, and the missing values and abnormal values are reconstructed and repaired to obtain the corresponding complete feature vector; Step 5: Divide the feature vectors obtained in step 3 and step 4 into a training set and a test set according to an 8:2 ratio; Step 6: Construct a state assessment model based on multi-dimensional data of high-voltage distribution cabinets using WOA-LSSVM; Step 7: Input the training set feature data into the state assessment model based on WOA-LSSVM high-voltage distribution cabinet multi-dimensional data, conduct training and learning of the feature data, grasp the internal change law, and obtain the size of the optimal key parameters of the LSSVM model; Step 8: Based on the optimal hyperparameters obtained in step 7, the state assessment model based on the multidimensional data of the WOA-LSSVM high-voltage distribution cabinet is rebuilt, and the test set data is input to verify the reliability of the model; Step 9: Using the WOA-LSSVM high-voltage distribution cabinet multi-dimensional data state assessment model, the different feature data sets collected by various sensors in real time are compared with the predicted values generated by the WOA-LSSVM model. If the error exceeds a certain range, an alarm is issued and manual inspection is performed to achieve real-time monitoring of abnormal conditions of the high-voltage distribution cabinet.
[0005] Preferably, in step 1, various important parameter indicators of the high-voltage distribution cabinet under normal operating conditions are determined according to the nameplate of the high-voltage distribution cabinet and manual experience, specifically including: the rated voltage U, rated current I, temperature of each component, humidity of each component, sound prints generated by each component, vibration generated by each component, characteristic gas concentration in the distribution cabinet and other specific variation ranges of the high-voltage distribution cabinet during normal operation.
[0006] Preferably, the series of sensors in step 2 include: a shell vibration sensor, a base plate vibration sensor, and a high-voltage distribution cabinet voiceprint sensor. The data information preprocessing of these two types of sensors is realized by the voiceprint recognition process; a high-voltage distribution cabinet temperature sensor, a high-voltage distribution cabinet humidity sensor, and a high-voltage distribution cabinet gas monitoring sensor.
[0007] Preferably, the step 3 specifically comprises the following steps: The first step is sound preprocessing, which is to pre-emphasize, frame, and window the sound, filter out the non-audio components in the sound, and then truncate it. The Hamming window function is as follows: (1); Where: I represents the number of frames, is the coefficient; The second step is fast Fourier transform (FFT), which deconstructs the preprocessed time domain signal into a frequency domain signal, making it suitable for training in a neural network and performing FFT processing on it: (2); Where: Represents the input sound signal, Indicates the number of FFT points; The third step is to take the square of the modulus of the signal in the frequency domain: (3); In the formula, Indicates signal No. frame; The fourth step is Mel filtering. The Mel filter consists of 40 triangular filters. The formula for converting frequency to Mel scale is: (4); In the formula, represents the final frequency, Indicates the actual frequency; The fifth step is to take the logarithm of the filtered result and calculate the formula of the logarithmic energy method of each filter output: (5); After the above steps, the Fbank feature is extracted. The extraction process of the Fbank feature is as follows: Figure 3 .
[0008] Preferably, in step 6, the principle of the WOA model is as follows: The role of the Whale Optimization Algorithm (WOA) is to iteratively optimize the key parameters of the neural network model. Its implementation process is divided into three steps: group surround the target, group attack the target, and search for the optimal target; (1) Group surrounds the target In the early stages of algorithm iteration, the whale group cannot lock on to the optimal target prey, so it first initializes the optimal position. (i, d represent the whale number and spatial position), and other groups move to the current position through equation (6); (6); Where: X(t+1), X(t), and Xbest(t) represent the position of the whale at the next moment, the current moment, and the optimal value at the current moment, respectively; is a linearly decreasing convergence factor between [2, 0], which can be realized by , , Represents a random number generated between [0, 1]; (2) Group attack targets The whale continuously updates its position horizontally and vertically, gradually surrounding the target group through equation (7); (7); Where: b is the spiral linear equation constant, which is 1; j is a random number in the range of [-1, 1]; when When , the group automatically determines whether to attack or shrink in the next moment, and the corresponding optimal value encirclement circle also changes continuously, and it updates its position according to the equal probability model, as shown in formula (8); (8); Where: represents an equal probability model, with values in the range [0, 1], and is the condition for the whale group to choose to surround or attack the prey; (3) Searching for the optimal target When the whale completes attacking the prey, based on the random whale position Xrang(t), according to formula (9), global continuous iteration is performed to find the optimal solution; (9).
[0009] Preferably, the principle of the LSSVM model in step 6 is as follows: Assume {x1, x2, ..., xn, yi}, x1∈Rn, x1 is the data set of the ith training sample, yi is the target value of the ith sample, and the total amount of the entire space is l, then the discriminant function is: (10); Where: is the weight vector; is the mapping function, b is the bias constant; The objective function of LSSVM is: (11); Where: is the penalty coefficient, represents the error variable; It is necessary to use the duality of the Lagrangian algorithm to assist in solving the problem, according to the KKT (Karush Kuhn-Tucker) condition, as shown in formula (12): (12); (13); The calculation can be obtained: (14); Where: , , , , is the key parameter of the Lagrange multiplier method, is the kernel function; The present invention adopts Gaussian radial basis function (RBF) as the kernel function of LSSVM, and substitutes the kernel function formula into equation (12) to obtain b and , we can get the LSSVM regression function as follows: (15); K is the Gaussian radial basis function (RBF) kernel function, and the penalty coefficient , kernel parameters is the key parameter of the LSSVM model. The present invention uses the WOA optimization algorithm to achieve the key parameters of the LSSVM model. , The search for excellence.
[0010] Preferably, the steps 7 and 8 are specifically as follows: (1) Normalize the feature vector processed in step 4. The specific method is shown in formula (16): (16); Where: is the normalized data; xmin and xmax are the minimum and maximum values of the sample data set respectively; x is the original sample data; (2) Set the maximum number of iterations, population size, and LSSVM model penalty coefficient of the WOA optimization model , kernel parameters The value range of , search speed range, and initialization population kernel parameters; (3) Use the training set data to experiment with the LSSVM network cleaning model, and calculate the fitness value of the particles, further update pbest and gbest, update the particle speed and displacement, and update the inertia weight and acceleration factor; (4) Determine whether the iteration termination condition is met (the particle fitness value tends to be flat or the number of particle iterations reaches the maximum). If the termination condition is met, output the optimal penalty coefficient , kernel parameters If the value of does not meet the judgment condition, return to step (3) and iterate again; (5) Based on the optimized optimal parameters, the WOA-LSSVM state assessment model of multi-dimensional data of high-voltage distribution cabinet is reconstructed.
[0011] The present invention provides a state assessment method for multi-dimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model, which has the following beneficial effects: (1) The present invention can quickly detect abnormal operating conditions of high-voltage distribution cabinets and use multi-dimensional data to effectively and accurately evaluate the fault status of high-voltage distribution cabinets; (2) Based on the fault information of the high-voltage distribution cabinet, the equipment can be inspected and maintained in a targeted manner, thereby reducing related safety accidents caused by faults. BRIEF DESCRIPTION OF THE DRAWINGS
[0012] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Figure 1 is a flow chart of the method of the present invention; Figure 2 This is a schematic diagram of Fbank voiceprint feature value extraction in the present invention; Figure 3 A process diagram for establishing a multidimensional data state assessment model of the present invention. DETAILED DESCRIPTION
[0013] like Figure 1 As shown, a state assessment method for multi-dimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model comprises the following steps: Step 1: According to the nameplate of the high-voltage distribution cabinet and manual experience, determine the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions; Step 2: According to the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions determined in step 1, a series of sensors are installed at corresponding positions of the high-voltage distribution cabinet to achieve the purpose of extracting various important parameter indicators; Step 3: Under normal operation of the high-voltage distribution cabinet, the data information collected by the voiceprint sensor and the vibration sensor installed in step 2 is processed using the Fbank technology to obtain the corresponding feature vector; Step 4: When the high-voltage distribution cabinet is in normal operation, the original data set collected by the computer monitoring system (voltage U, current I, temperature of each component, humidity of each component, and change range of characteristic gas concentration in the distribution cabinet during operation of the high-voltage distribution cabinet) is collected, and the original characteristic data set is cleaned by using the machine learning model, and the missing values and abnormal values are reconstructed and repaired to obtain the corresponding complete feature vector; Step 5: Divide the feature vectors obtained in step 3 and step 4 into a training set and a test set according to an 8:2 ratio; Step 6: Construct a state assessment model based on multi-dimensional data of high-voltage distribution cabinets using WOA-LSSVM; Step 7: Input the training set feature data into the state assessment model based on WOA-LSSVM high-voltage distribution cabinet multi-dimensional data, conduct training and learning of the feature data, grasp the internal change law, and obtain the size of the optimal key parameters of the LSSVM model; Step 8: Based on the optimal hyperparameters obtained in step 7, the state assessment model based on the multidimensional data of the WOA-LSSVM high-voltage distribution cabinet is rebuilt, and the test set data is input to verify the reliability of the model; Step 9: Using the WOA-LSSVM high-voltage distribution cabinet multi-dimensional data state assessment model, the different feature data sets collected by various sensors in real time are compared with the predicted values generated by the WOA-LSSVM model. If the error exceeds a certain range, an alarm is issued and manual inspection is performed to achieve real-time monitoring of abnormal conditions of the high-voltage distribution cabinet.
[0014] Preferably, in step 1, various important parameter indicators of the high-voltage distribution cabinet under normal operating conditions are determined according to the nameplate of the high-voltage distribution cabinet and manual experience, specifically including: the rated voltage U, rated current I, temperature of each component, humidity of each component, sound prints generated by each component, vibration generated by each component, characteristic gas concentration in the distribution cabinet and other specific variation ranges of the high-voltage distribution cabinet during normal operation.
[0015] Preferably, the series of sensors in step 2 include: a shell vibration sensor, a base plate vibration sensor, and a high-voltage distribution cabinet voiceprint sensor. The data information preprocessing of these two types of sensors is realized by the voiceprint recognition process; a high-voltage distribution cabinet temperature sensor, a high-voltage distribution cabinet humidity sensor, and a high-voltage distribution cabinet gas monitoring sensor.
[0016] Preferably, Figure 2 As shown, the step 3 specifically includes the following steps: The first step is sound preprocessing, which is to pre-emphasize, frame, and window the sound, filter out the non-audio components in the sound, and then truncate it. The Hamming window function is as follows: (1); Where: I represents the number of frames, is the coefficient; The second step is fast Fourier transform (FFT), which deconstructs the preprocessed time domain signal into a frequency domain signal, making it suitable for training in a neural network and performing FFT processing on it: (2); Where: Represents the input sound signal, Indicates the number of FFT points; The third step is to take the square of the modulus of the signal in the frequency domain: (3); In the formula, Indicates signal No. frame; The fourth step is Mel filtering. The Mel filter consists of 40 triangular filters. The formula for converting frequency to Mel scale is: (4); In the formula, represents the final frequency, Indicates the actual frequency; The fifth step is to take the logarithm of the filtered result and calculate the formula of the logarithmic energy method of each filter output: (5); After the above steps, the Fbank feature is extracted. The extraction process of the Fbank feature is as follows: Figure 3 .
[0017] Preferably, in step 6, the principle of the WOA model is as follows: The role of the Whale Optimization Algorithm (WOA) is to iteratively optimize the key parameters of the neural network model. Its implementation process is divided into three steps: group surround the target, group attack the target, and search for the optimal target; (1) Group surrounds the target In the early stages of algorithm iteration, the whale group cannot lock on to the optimal target prey, so it first initializes the optimal position. (i, d represent the whale number and spatial position), and other groups move to the current position through equation (6); (6); Where: X(t+1), X(t), and Xbest(t) represent the position of the whale at the next moment, the current moment, and the optimal value at the current moment, respectively; is a linearly decreasing convergence factor between [2, 0], which can be realized by , , Represents a random number generated between [0, 1]; (2) Group attack targets The whale continuously updates its position horizontally and vertically, gradually surrounding the target group through equation (7); (7); Where: b is the spiral linear equation constant, which is 1; j is a random number in the range of [-1, 1]; when When , the group automatically determines whether to attack or shrink in the next moment, and the corresponding optimal value encirclement circle also changes continuously, and it updates its position according to the equal probability model, as shown in formula (8); (8); Where: represents an equal probability model, with values in the range [0, 1], and is the condition for the whale group to choose to surround or attack the prey; (3) Searching for the optimal target When the whale completes attacking the prey, based on the random whale position Xrang(t), according to formula (9), global continuous iteration is performed to find the optimal solution; (9).
[0018] Preferably, the principle of the LSSVM model in step 6 is as follows: Assume {x1, x2, ..., xn, yi}, x1∈Rn, x1 is the data set of the ith training sample, yi is the target value of the ith sample, and the total amount of the entire space is l, then the discriminant function is: (10); Where: is the weight vector; is the mapping function, b is the bias constant; The objective function of LSSVM is: (11); Where: is the penalty coefficient, represents the error variable; It is necessary to use the duality of the Lagrangian algorithm to assist in solving the problem, according to the KKT (Karush Kuhn-Tucker) condition, as shown in formula (12): (12); (13); The calculation can be obtained: (14); Where: , , , , is the key parameter of the Lagrange multiplier method, is the kernel function; The present invention adopts Gaussian radial basis function (RBF) as the kernel function of LSSVM, and substitutes the kernel function formula into equation (12) to obtain b and , we can get the LSSVM regression function as follows: (15); K is the Gaussian radial basis function (RBF) kernel function, and the penalty coefficient , kernel parameters is the key parameter of the LSSVM model. The present invention uses the WOA optimization algorithm to achieve the key parameters of the LSSVM model. , The search for excellence.
[0019] Preferably, Figure 3 As shown, the steps 7 and 8 are specifically as follows: (1) Normalize the feature vector processed in step 4. The specific method is shown in formula (16): (16); Where: is the normalized data; xmin and xmax are the minimum and maximum values of the sample data set respectively; x is the original sample data; (2) Set the maximum number of iterations, population size, and LSSVM model penalty coefficient of the WOA optimization model , kernel parameters The value range of , search speed range, and initialization population kernel parameters; (3) Use the training set data to experiment with the LSSVM network cleaning model, and calculate the fitness value of the particles, further update pbest and gbest, update the particle speed and displacement, and update the inertia weight and acceleration factor; (4) Determine whether the iteration termination condition is met (the particle fitness value tends to be flat or the number of particle iterations reaches the maximum). If the termination condition is met, output the optimal penalty coefficient , kernel parameters If the value of does not meet the judgment condition, return to step (3) and iterate again; (5) Based on the optimized optimal parameters, the WOA-LSSVM state assessment model of multi-dimensional data of high-voltage distribution cabinet is reconstructed.
[0020] When in use, the main hardware includes: 2 voiceprint collectors, installed in the high-voltage power distribution cabinet and components respectively; 2 vibration sensors, installed in the shell and bottom plate respectively; the voiceprint sensor uses the ZKHY-IVOD-013 industrial diagnostic pickup, with a collection frequency of 20HZ-20KHZ; the vibration sensor uses the SE830 piezoelectric vibration sensor; the voiceprint collection edge gateway terminal integrated machine ZKHY-IVH-002, the collection card has 8 audio inputs, and the input voltage is DC 9~36V. The voiceprint server uses the cloud engine server ZKHY-VCES-2000. Then perform the following operations: Voice signal reading: Use the voiceprint sensor to collect the voiceprint signal of the high-voltage distribution cabinet during operation at a sampling rate of 16kHz and a sampling accuracy of 24bit, and use the audio processing library to read the input voice signal.
[0021] Process the sound samples; Preprocessing: Perform pre-emphasis on the audio signal, and set the pre-emphasis coefficient to 0.97. Then, divide the audio signal into frames, with each frame length of 16ms and a frame shift of 8ms. Finally, perform windowing on each frame, and select the Hamming window as the window function.
[0022] Fast Fourier Transform: Perform fast Fourier transform on each frame of speech signal to obtain its spectrum representation. The number of fast Fourier transform points is set to 512, and the window function selects the Hamming window.
[0023] Mel filtering: Design a Mel filter bank with 24 filters. The filter frequency range covers the entire spectrum range of the speech signal. Filter the spectrum obtained by fast Fourier transform through the Mel filter bank to obtain the Mel spectrum.
[0024] Take logarithm: Perform logarithmic transformation on the Mel spectrum to obtain the logarithmic Mel spectrum. The logarithmic transformation uses the natural logarithm function.
[0025] Feature extraction: The logarithmic Mel spectrum is extracted as Fbank features. The extracted feature parameters include the logarithmic energy value of each filter channel.
[0026] The present invention proposes a multi-dimensional data status evaluation method for a high-voltage distribution cabinet based on a WOA-LSSVM model. Firstly, various important parameter indicators of the high-voltage distribution cabinet under normal operating conditions are determined according to the nameplate of the high-voltage distribution cabinet and manual experience, and a series of sensors are installed at corresponding positions of the high-voltage distribution cabinet according to such parameter indicators to obtain corresponding feature data sets; then, the collected data information is processed using Fbank and a machine learning model to obtain corresponding feature vectors, and the feature vectors are divided into a training set and a test set according to an 8:2 ratio; finally, a status evaluation model for multi-dimensional data of a high-voltage distribution cabinet based on the WOA-LSSVM is constructed, and training and learning of feature data are performed using training set data to master the internal variation law, and the size of the optimal key parameters of the LSSVM model is obtained, and the test set data is input to verify the reliability of the model, and different feature data sets collected by various sensors in real time are compared with the predicted values generated by the WOA-LSSVM model. If the error exceeds a certain range, an alarm is issued. The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.
Claims
1. A method for evaluating the state of multidimensional data of high-voltage distribution cabinets based on the WOA-LSSVM model, characterized in that: The following steps are involved: Step 1: According to the nameplate of the high-voltage distribution cabinet and manual experience, determine the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions; Step 2: According to the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions determined in step 1, a series of sensors are installed at corresponding positions of the high-voltage distribution cabinet to achieve the purpose of extracting various important parameter indicators; Step 3: Under normal operation of the high-voltage distribution cabinet, the data information collected by the voiceprint sensor and the vibration sensor installed in step 2 is processed using the Fbank technology to obtain the corresponding feature vector; Step 4: Under normal operation of the high-voltage distribution cabinet, collect the original data set collected by the computer monitoring system, and use the machine learning model to clean the original feature data set, reconstruct and repair missing values and outliers, and obtain the corresponding complete feature vector; Step 5: Divide the feature vectors obtained in step 3 and step 4 into a training set and a test set according to an 8:2 ratio; Step 6: Construct a state assessment model based on multi-dimensional data of high-voltage distribution cabinets using WOA-LSSVM; Step 7: Input the training set feature data into the state assessment model based on WOA-LSSVM high-voltage distribution cabinet multi-dimensional data, conduct training and learning of the feature data, grasp the internal change law, and obtain the size of the optimal key parameters of the LSSVM model; Step 8: Based on the optimal hyperparameters obtained in step 7, the state assessment model based on the multidimensional data of the WOA-LSSVM high-voltage distribution cabinet is rebuilt, and the test set data is input to verify the reliability of the model; Step 9: Using the WOA-LSSVM high-voltage distribution cabinet multi-dimensional data state assessment model, the different feature data sets collected by various sensors in real time are compared with the predicted values generated by the WOA-LSSVM model. If the error exceeds a certain range, an alarm is issued and manual inspection is performed to achieve real-time monitoring of abnormal conditions of the high-voltage distribution cabinet.
2. According to claim 1, a method for evaluating the state of multidimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model, characterized in that: In the step 1, according to the nameplate of the high-voltage distribution cabinet and manual experience, the important parameter indicators of the high-voltage distribution cabinet under normal operating conditions are determined, specifically including: the rated voltage U, rated current I, temperature of each component, humidity of each component, sound prints generated by each component, vibration generated by each component, characteristic gas concentration in the distribution cabinet and other specific variation ranges during normal operation of the high-voltage distribution cabinet.
3. According to the state assessment method of multidimensional data of high-voltage distribution cabinet based on WOA-LSSVM model in claim 1, it is characterized in that: The series of sensors in step 2 include: shell vibration sensor, base plate vibration sensor, high-voltage distribution cabinet voiceprint sensor. The data information preprocessing of these two types of sensors is realized by the voiceprint recognition process; high-voltage distribution cabinet temperature sensor, high-voltage distribution cabinet humidity sensor and high-voltage distribution cabinet gas monitoring sensor.
4. According to the state assessment method of multidimensional data of high-voltage distribution cabinet based on WOA-LSSVM model in claim 1, it is characterized in that: The step 3 specifically includes the following steps: The first step is sound preprocessing, which is to pre-emphasize, frame, and window the sound, filter out the non-audio components in the sound, and then truncate it. The Hamming window function is as follows: (1); Where: I represents the number of frames, is the coefficient; The second step is fast Fourier transform, which deconstructs the preprocessed time domain signal into a frequency domain signal, making it suitable for training in a neural network and performing FFT processing on it: (2); Where: Represents the input sound signal, Indicates the number of FFT points; The third step is to take the square of the modulus of the signal in the frequency domain: (3); In the formula, Indicates signal No. frame; The fourth step is Mel filtering. The Mel filter consists of 40 triangular filters. The formula for converting frequency to Mel scale is: (4); In the formula, represents the final frequency, Indicates the actual frequency; The fifth step is to take the logarithm of the filtered result and calculate the formula of the logarithmic energy method of each filter output: (5); After the above steps, the Fbank features are extracted.
5. According to claim 1, a method for evaluating the state of multi-dimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model, characterized in that: In step 6, the principle of the WOA model is as follows: The role of the whale optimization algorithm is to iteratively optimize the key parameters of the neural network model. Its implementation process is divided into three steps: group surround the target, group attack the target, and search for the optimal target; (1) Group surrounds the target In the early stages of algorithm iteration, the whale group cannot lock on to the optimal target prey, so it first initializes the optimal position. , other groups move to the current position through equation (6); (6); Where: X(t+1), X(t), and Xbest(t) represent the position of the whale at the next moment, the current moment, and the optimal value at the current moment, respectively; is a linearly decreasing convergence factor between [2, 0], which can be realized by , , Represents a random number generated between [0, 1]; (2) Group attack targets The whale continuously updates its position horizontally and vertically, gradually surrounding the target group through equation (7); (7); Where: b is the spiral linear equation constant, which is 1; j is a random number in the range of [-1, 1]; when When , the group automatically determines whether to attack or shrink in the next moment, and the corresponding optimal value encirclement circle also changes continuously, and it updates its position according to the equal probability model, as shown in formula (8); (8); Where: represents an equal probability model, with values in the range [0, 1], and is the condition for the whale group to choose whether to surround or attack the prey; (3) Searching for the optimal target When the whale completes attacking the prey, based on the random whale position Xrang(t), according to formula (9), global continuous iteration is performed to find the optimal solution; (9)。 6. According to claim 1, a method for evaluating the state of multi-dimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model, characterized in that: The principle of the LSSVM model in step 6 is as follows: Assume {x1, x2, ..., xn, yi}, x1∈Rn, x1 is the data set of the ith training sample, yi is the target value of the ith sample, and the total amount of the entire space is l, then the discriminant function is: (10); Where: is the weight vector; is the mapping function, b is the bias constant; The objective function of LSSVM is: (11); Where: is the penalty coefficient, represents the error variable; It is necessary to use the duality of the Lagrangian algorithm to assist in solving the problem. According to the KKT condition, as shown in formula (12): (12); (13); The calculation can be obtained: (14); Where: , , , , is the key parameter of the Lagrange multiplier method, is the kernel function; The Gaussian radial basis function is used as the kernel function of LSSVM. The kernel function formula is substituted into equation (12) and the solution is obtained by combining b and , we can get the LSSVM regression function as follows: (15); K is the Gaussian radial basis function kernel function, and the penalty coefficient , kernel parameters The key parameters of the LSSVM model are optimized using the WOA algorithm. , The search for excellence.
7. According to claim 1, a method for evaluating the state of multi-dimensional data of a high-voltage distribution cabinet based on a WOA-LSSVM model, characterized in that: The steps 7 and 8 are as follows: (1) Normalize the feature vector processed in step 4. The specific method is shown in formula (16): (16); Where: is the normalized data; xmin and xmax are the minimum and maximum values of the sample data set respectively; x is the original sample data; (2) Set the maximum number of iterations, population size, and LSSVM model penalty coefficient of the WOA optimization model , kernel parameters The value range of , search speed range, and initialization population kernel parameters; (3) Use the training set data to experiment with the LSSVM network cleaning model, and calculate the fitness value of the particles, further update pbest and gbest, update the particle speed and displacement, and update the inertia weight and acceleration factor; (4) Determine whether the iteration termination condition is met (the particle fitness value tends to be flat or the number of particle iterations reaches the maximum). If the termination condition is met, output the optimal penalty coefficient , kernel parameters If the value of does not meet the judgment condition, return to step (3) and iterate again; (5) Based on the optimized optimal parameters, the WOA-LSSVM state assessment model of multi-dimensional data of high-voltage distribution cabinet is reconstructed.
