Small current grounding system fault line selection method of improved bat algorithm optimized support vector machine based on feature fusion

Abstract

The invention discloses a small current grounding system fault line selection method based on a feature fusion improved bat algorithm optimized support vector machine, and belongs to the field of medium voltage power distribution system fault line selection. The method comprises the following steps: firstly, processing a line zero-sequence current signal by using a denoising method in which S transformation threshold filtering and a time-frequency filter based on time-frequency spectrum distribution are combined, then performing line selection by taking S transformation energy relative entropy and singular entropy as double characteristic quantities by using a characteristic fusion technology, and then optimizing a support vector machine (SVM) by using an improved bat algorithm (IBA). And an IBA-SVM optimization classifier is constructed to carry out classification processing on the fault feature data set, and finally identification of the fault line is realized. Simulation and experimental data verification show that the method has high accuracy and practicability.

Description

Technical Field

[0001] The present invention relates to the field of medium voltage distribution network fault protection, and in particular to a method for fault line selection of a small current grounding system based on an improved bat algorithm and optimized support vector machine based on feature fusion. Background Art

[0002] Single-phase grounding faults account for more than 80% of the low-current grounding system. When a single-phase grounding fault occurs in the system, the line voltage remains symmetrical, so it will not affect the load, but the non-fault phase voltage will rise to the line voltage. If the fault line is not identified and isolated in time, more areas will fail, and eventually the entire system will be transformed from faulty operation to fault power outage. Therefore, how to improve the accuracy and efficiency of fault line selection is of great significance to ensure the safe and reliable operation of the low-current grounding system.

[0003] Existing line selection methods are roughly divided into steady-state methods, transient methods, and comprehensive line selection methods. Line selection methods based on steady-state information include zero-sequence admittance method, amplitude-phase method, fifth harmonic method, residual current increment method, etc. However, since the steady-state characteristic quantity is affected by factors such as arc suppression coils and grounding resistance, the line selection accuracy is reduced. The fault characteristics contained in the transient process are usually richer than those in the steady-state process, and the current amplitude is larger, so it is more advantageous. Specifically, it includes the first half-wave method, wavelet analysis method, Prony method, etc. However, the above methods are easily affected by various factors such as signal noise and optimal parameters, and most of them use a single criterion, have poor anti-interference ability, and have low line selection accuracy.

[0004] In recent years, multi-criteria fusion line selection methods based on artificial intelligence theory have become a new trend. This method combines fuzzy theory with neural networks, but the adjustment of the weight coefficient of this method needs to consider factors such as the time when the fault occurs and the size of the transition resistance, which may interfere with the line selection effect. Clustering algorithms, neural networks, and Petri nets have the disadvantages of being easy to fall into local optimality, being sensitive to randomly mixed interference data, taking too long to train the model, and having certain requirements on the number of training samples. SVM, which is trained with small samples, can simplify high-dimensional space with its unique mechanism, thereby reducing classification pressure and improving the ability to filter invalid samples. SVM has good comprehensive performance and high adaptability to small sample problems, but there are some defects in the selection of parameters, so it is necessary to introduce optimization algorithms to optimize its key parameters. Summary of the invention

[0005] In order to solve the problems mentioned in the above background technology, the present invention proposes a fault line selection method for a small current grounding system based on an improved bat algorithm optimized support vector machine based on feature fusion.

[0006] The present invention achieves the above technical objectives and technical effects through the following technical solutions: a method for fault line selection in a small current grounding system based on feature fusion-based improved bat algorithm optimized support vector machine, which comprises the following steps: Step 1: monitor the neutral point voltage of the distribution network. When the zero-sequence voltage exceeds the specified threshold, it is determined that a fault has occurred in the system and the fault line selection is started; Step 2: Extract the zero-sequence current signal of the line under each fault condition, take the cycle from 1 / 4 before the fault to 3 / 4 after the fault, perform S transformation and signal preprocessing, take the modulus time-frequency matrix of the complex time-frequency matrix obtained after the transformation to calculate the energy relative entropy and singular entropy and fuse them, use the fused feature quantity as the fault feature vector of the IBA-SVM optimization classifier, construct a feature sample set and normalize it to ensure that all data are in the same scale; Step 3: Select a part of the data that can contain various fault conditions from the fusion samples as the training sample set of SVM, and use the other part as the test sample set to verify the performance of the model; Step 4: Using the search capability of IBA, find the parameter combination that makes the SVM classification performance the best, apply the optimized parameters to SVM and establish the IBA-SVM optimized classifier, and use the training sample set for training; Step 5: Input the test samples into the trained IBA-SVM model. The model classifies the test samples according to the input feature vector and completes the fault line selection based on different line labels.

[0007] Furthermore, the specific steps of using S transform to denoise in step 2 are: Firstly, a discrete sampling sequence of a set of continuous time-domain current signals is input, and a two-dimensional time-frequency complex matrix is ​​obtained by S transform. Then, the signal is processed by threshold filtering method and time-frequency filter based on time-frequency spectrum distribution. Finally, the one-dimensional time-domain current signal with noise removed is obtained by inverse S transform.

[0008] Furthermore, the calculation method of the energy relative entropy in step 2 is: 1) According to the following definition Frequency Outgoing lines (sampling time points are ) is: 2) According to the following formula, sum the transient current energy of each outgoing line at this frequency, and get The transient current energy and : In the formula, Indicates the number of lines; 3) According to the following formula, we can get Frequency, The ratio of the energy of the outgoing line to the total energy is : 4) Calculate each outgoing line according to the following formula Relative to a feeder S-transform energy relative entropy : In the formula, ; 5) Calculate the output line according to the following formula Relative entropy relative to the total energy of the rest of the circuit :

[0009] Furthermore, the calculation method of the singular entropy in step 2 is: 1) Any Order Matrix The singular values ​​of can be decomposed into: In the formula, and They are Step and orthogonal matrix, Yes Yes Matrix A diagonal matrix with diagonal elements For the matrix The singular values ​​of 2) According to the singular value decomposition, a rank of Order Matrix express A rank 1 The sum of the submatrices of order, each of which consists of two eigenvectors (one from each and ) and the weights to get: In the formula, for rank, , They are and No. Column singular value vectors, To include and submatrix of ; 3) The modulus time-frequency matrix obtained after S-transformation of the zero-sequence current signal of each line is decomposed by singular value to obtain different singular values , calculate the energy of the obtained singular value and normalize it, recorded as , construct the singular entropy expression as:

[0010] Furthermore, the method for finding the optimal parameters of SVM classification performance by IBA in step 4 is: 1) Set the basic parameters of the bat population and the maximum number of iterations. Each solution obtained corresponds to a set of SVM and ; 2) The root mean square error function is selected as the fitness function, the initial fitness value of the individual bat is calculated, the solution of the optimal bat location is selected, and the speed and position are updated. The selected fitness function is: 3) Generate pulse intensity through local optimization, generate new solutions based on fitness value, explore near the global optimal solution, and obtain possible local new solutions; 4) Record the current optimal solution when a sudden change occurs during the optimization process, expand the search range, and repeat steps 2 to 4 until the maximum number of iterations is reached or the preset conditions are met; 5) The parameters corresponding to the maximum fitness value during the iteration process are output as the optimal parameters of SVM.

[0011] After adopting the above method, the present invention has the following advantages compared with the existing single-phase grounding fault line selection technology: 1. A filtering denoising method combining S-transform threshold filtering with a time-frequency filter based on time-frequency spectrum distribution is used to extract the S-transform energy relative entropy and singular entropy of the zero-sequence current of each line as fault feature quantities. Using feature fusion technology, the information of the two feature quantities is integrated for line selection, which makes up for the deficiency of using only a single feature quantity for fault line selection and reduces the possibility of misjudgment of line selection.

[0012] 2. The improved bat algorithm is used to optimize the selection of support vector machine penalty parameters and kernel parameters, and feature fusion technology is combined with IBA-SVM optimization classifier for fault line selection. This method has good superiority and accuracy in fault line selection. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 It is a flow chart for line selection for single-phase grounding fault in a small current grounding system.

[0014] Figure 2 It is the structure diagram of distribution network simulation.

[0015] Figure 3 This is the line selection result diagram of the unoptimized SVM test set.

[0016] Figure 4 This is the IBA-SVM test set line selection result diagram. DETAILED DESCRIPTION

[0017] In order to better understand the present invention, the content of the present invention is further described below in conjunction with the accompanying drawings and examples.

[0018] A fault line selection method for small current grounding system based on feature fusion and improved bat algorithm to optimize support vector machine Figure 1 As shown, the specific process includes the following steps: Step 1: monitor the neutral point voltage of the distribution network. When the zero-sequence voltage exceeds the specified threshold, it is determined that a fault has occurred in the system and the fault line selection is started; Step 2: Extract the zero-sequence current signal of the line under each fault condition, take the cycle from 1 / 4 before the fault to 3 / 4 after the fault, perform S transformation and signal preprocessing, take the modulus time-frequency matrix of the complex time-frequency matrix obtained after the transformation to calculate the energy relative entropy and singular entropy and fuse them, use the fused feature quantity as the fault feature vector of the IBA-SVM optimization classifier, construct a feature sample set and normalize it to ensure that all data are in the same scale; Step 3: Select a part of the data that can contain various fault conditions from the fusion samples as the training sample set of SVM, and use the other part as the test sample set to verify the performance of the model; Step 4: Using the search capability of IBA, find the parameter combination that makes the SVM classification performance the best, apply the optimized parameters to SVM and establish the IBA-SVM optimized classifier, and use the training sample set for training; Step 5: Input the test samples into the trained IBA-SVM model. The model classifies the test samples according to the input feature vector and completes the fault line selection based on different line labels.

[0019] Furthermore, the specific steps of using S transform to denoise in step 2 are: Firstly, a discrete sampling sequence of a set of continuous time-domain current signals is input, and a two-dimensional time-frequency complex matrix is ​​obtained by S transform. Then, the signal is processed by threshold filtering method and time-frequency filter based on time-frequency spectrum distribution. Finally, the one-dimensional time-domain current signal with noise removed is obtained by inverse S transform.

[0020] Furthermore, the calculation method of the energy relative entropy in step 2 is: 1) According to the following definition Frequency Outgoing lines (sampling time points are ) is: 2) According to the following formula, sum the transient current energy of each outgoing line at this frequency, and get The transient current energy and : In the formula, Indicates the number of lines; 3) According to the following formula, we can get Frequency, The ratio of the energy of the outgoing line to the total energy is : 4) Calculate each outgoing line according to the following formula Relative to a feeder S-transform energy relative entropy : In the formula, ; 5) Calculate the output line according to the following formula Relative entropy relative to the total energy of the rest of the circuit :

[0021] Furthermore, the calculation method of the singular entropy in step 2 is: 1) Any Order Matrix The singular values ​​of can be decomposed into: In the formula, and They are Step and orthogonal matrix, Yes Yes Matrix A diagonal matrix with diagonal elements For the matrix The singular values ​​of 2) According to the singular value decomposition, a rank of Order Matrix express A rank 1 The sum of the submatrices of order, each of which consists of two eigenvectors (one from each and ) and the weights to get: In the formula, for rank, , They are and No. Column singular value vectors, To include and submatrix of ; 3) The modulus time-frequency matrix obtained after S-transformation of the zero-sequence current signal of each line is decomposed by singular value to obtain different singular values , calculate the energy of the obtained singular value and normalize it, recorded as , construct the singular entropy expression as:

[0022] Furthermore, the method for finding the optimal parameters of SVM classification performance by IBA in step 4 is: 1) Set the basic parameters of the bat population and the maximum number of iterations. Each solution obtained corresponds to a set of SVM and ; 2) The root mean square error function is selected as the fitness function, the initial fitness value of the individual bat is calculated, the solution of the optimal bat location is selected, and the speed and position are updated. The selected fitness function is: 3) Generate pulse intensity through local optimization, generate new solutions based on fitness value, explore near the global optimal solution, and obtain possible local new solutions; 4) Record the current optimal solution when a sudden change occurs during the optimization process, expand the search range, and repeat steps 2 to 4 until the maximum number of iterations is reached or the preset conditions are met; 5) The parameters corresponding to the maximum fitness value during the iteration process are output as the optimal parameters of SVM.

[0023] Simulation Verification

[0024] exist Figure 2 A large number of simulation experiments were carried out in the simulation model shown in the figure, and a total of 1120 sets of fault feature data were obtained. From the total feature data, 900 sets of feature data containing all fault conditions were selected to form the fault training sample set, and the remaining 220 sets of feature data were used as the fault test sample set. The distribution of the fault feature samples is shown in Table 1, where H represents the relative entropy of S-transform energy and Q represents the singular entropy of S-transform.

[0025] Table 1. Distribution of feature sample data Neutral point grounding method Fault type Fault line Fault location / % Transition resistance / Ω Initial phase angle / ° Neutral point is not grounded, but grounded via arc suppression coil Grounding through a resistor L1~L4 0~100 50~5000 0~90 800 groups Neutral point is not grounded, but grounded via arc suppression coil Metal grounding L1~L4 0~100 0~90 160 groups Neutral point is not grounded, but grounded via arc suppression coil High resistance grounding L1~L4 0~100 0~90 160 groups Based on the above fault feature data samples, 225 groups of feature data containing different faults were selected from the four lines as training sample sets, and the remaining 220 groups were used as test sample sets. The category label 1 represents that line L1 has a fault, the category label 2 represents that line L2 has a fault, and the category 3 and category 4 are similar. "*" represents the line selection result of the classifier.

[0026] Figure 3 This is the line selection test result of the SVM classifier without any optimization. The line selection accuracy is 92.73%. It can be found that no matter which line fails, the line selection error will occur. In view of the shortcomings of the SVM classifier, the present invention uses the improved bat algorithm IBA to optimize the SVM classifier, constructs the IBA-SVM optimized classifier model, and tests its fault line selection accuracy. The fault line selection of this model is shown in Figure 2. Figure 4 As shown. Figure 4 It can be seen that the IBA-SVM optimized classifier made misjudgments when two cable lines failed, and the overall line selection accuracy reached 98.18%. Compared with the unoptimized SVM classifier, the optimized SVM line selection accuracy was greatly improved. It can be seen that the method proposed in the present invention has significantly improved the accuracy and reliability of the line selection results.

[0027] In order to further verify the practical value of the present invention, a low-voltage equivalent experimental platform was built to simulate the fault characteristics of the zero-sequence current of each line when a single-phase grounding fault occurred in a 10kV distribution network, and the effectiveness of the present invention was verified.

[0028] Set line 1 to have a grounding fault through a transition resistor, the sampling frequency is 12.5kHz, the transition resistor value is changed, and other conditions remain unchanged. The energy relative entropy and singular entropy of each line after S transformation are calculated and normalized by the method proposed in the present invention, and the calculated fault feature quantity is used to form a fusion dual feature quantity, which is input into the constructed IBA-SVM optimization classifier for fault line selection identification. The identification results are shown in Table 2. It can be seen from the table that the identification results of several groups of data are correct, and the judgment results are all that line 1 has a fault, which can prove that the present invention has a certain practical application ability.

[0029] Table 2 Fault line identification results Transition resistance / Ω [H1,H2,H3,H4,Q1,Q2,Q3,Q4] Identify faulty lines 500 [0.9704,0.1018,0.0707,0.0296,0.6717,0.3283,0.3546,0.3808] Line 1 1000 [0.9232,0.3821,0.2464,0.0768,0.6528,0.3471,0.3693,0.3794] Line 1 2000 [0.8095,0.3284,0.1905,0.2067,0.6299,0.3701,0.3875,0.4049] Line 1 5000 [0.7696,0.7148,0.5653,0.2304,0.5834,0.5144,0.4323,0.4168] Line 1 The above description and embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for fault line selection in a small current grounding system based on feature fusion and improved bat algorithm to optimize support vector machine, characterized in that: The steps include: Step 1: monitor the neutral point voltage of the distribution network. When the zero-sequence voltage exceeds the specified threshold, it is determined that a fault has occurred in the system and the fault line selection is started; Step 2: Extract the zero-sequence current signal of the line under each fault condition, take the cycle from 1 / 4 before the fault to 3 / 4 after the fault, perform S transformation and signal preprocessing, take the modulus time-frequency matrix of the complex time-frequency matrix obtained after the transformation to calculate the energy relative entropy and singular entropy and fuse them, use the fused feature quantity as the fault feature vector of the IBA-SVM optimization classifier, construct a feature sample set and normalize it to ensure that all data are in the same scale; Step 3: Select a part of the data that can contain various fault conditions from the fusion samples as the training sample set of SVM, and use the other part as the test sample set to verify the performance of the model; Step 4: Using the search capability of IBA, find the parameter combination that makes the SVM classification performance the best, apply the optimized parameters to SVM and establish the IBA-SVM optimized classifier, and use the training sample set for training; Step 5: Input the test samples into the trained IBA-SVM model. The model classifies the test samples according to the input feature vector and completes the fault line selection based on different line labels.

2. The method for fault line selection in a small current grounding system based on feature fusion and improved bat algorithm optimized support vector machine according to claim 1 is characterized in that: The specific steps of using S transform to denoise in step 2 are as follows: Firstly, a discrete sampling sequence of a set of continuous time-domain current signals is input, and a two-dimensional time-frequency complex matrix is ​​obtained by S transform. Then, the signal is processed by threshold filtering method and time-frequency filter based on time-frequency spectrum distribution. Finally, the one-dimensional time-domain current signal with noise removed is obtained by inverse S transform.

3. The method for fault line selection in a small current grounding system based on feature fusion and improved bat algorithm optimized support vector machine according to claim 1 is characterized in that: The calculation method of the energy relative entropy in step 2 is: 1) According to the following definition Frequency Outgoing lines (sampling time points are ) is: 2) According to the following formula, sum the transient current energy of each outgoing line at this frequency, and get The transient current energy and : In the formula, Indicates the number of lines; 3) According to the following formula, we can get Frequency, The ratio of the energy of the outgoing line to the total energy is : 4) Calculate each outgoing line according to the following formula Relative to a feeder S-transform energy relative entropy : In the formula, ; 5) Calculate the output line according to the following formula Relative entropy relative to the total energy of the rest of the circuit :

4. The method for fault line selection in a small current grounding system based on feature fusion and improved bat algorithm optimization support vector machine according to claim 1 is characterized in that: The calculation method of singular entropy in step 2 is: 1) Any Order Matrix The singular values ​​of can be decomposed into: In the formula, and They are Step and orthogonal matrix, Yes Yes Matrix A diagonal matrix with diagonal elements For the matrix The singular values ​​of 2) According to the singular value decomposition, a rank of Order Matrix express A rank 1 The sum of the submatrices of order, each submatrix consists of two eigenvectors (one from each and ) and the weights to get: In the formula, for rank, , They are and No. Column singular value vectors, To include and submatrix of ; 3) The modulus time-frequency matrix obtained after S-transformation of the zero-sequence current signal of each line is decomposed by singular value to obtain different singular values , calculate the energy of the obtained singular value and normalize it, recorded as , construct the singular entropy expression as:

5. The method for fault line selection in a small current grounding system based on feature fusion and improved bat algorithm optimized support vector machine according to claim 1 is characterized in that: The method for finding the optimal parameters for SVM classification performance through IBA in step 4 is: 1) Set the basic parameters of the bat population and the maximum number of iterations. Each solution obtained corresponds to a set of SVM and ; 2) The root mean square error function is selected as the fitness function, the initial fitness value of the individual bat is calculated, the solution of the optimal bat location is selected, and the speed and position are updated. The selected fitness function is: 3) Generate pulse intensity through local optimization, generate new solutions based on fitness value, explore near the global optimal solution, and obtain possible local new solutions; 4) Record the current optimal solution when a sudden change occurs during the optimization process, expand the search range, and repeat steps 2 to 4 until the maximum number of iterations is reached or the preset conditions are met; 5) The parameters corresponding to the maximum fitness value during the iteration process are output as the optimal parameters of SVM.

Images