GIS partial discharge fault identification method based on improved mountaineering optimization algorithm

Abstract

The invention discloses a GIS partial discharge fault identification method based on an improved mountaineering optimization algorithm. The method comprises the following steps: 1) collecting four typical fault signals of partial discharge; 2) extracting a mean value, a root mean square, a variance, skewness, a kurtosis index, a pulse index, a peak value index, a margin index and a waveform index of the partial discharge signal as feature vectors; 3) using a refraction reverse learning and Cauchy variation strategy to improve a mountaineering optimization algorithm; 4) in the training set, using an improved mountaineering optimization algorithm to optimize an input weight and a hidden layer threshold of the extreme learning machine, and after the maximum number of iterations is reached, inputting a coordinate with an optimal solution as an optimal parameter combination into an ELM model to establish an SCMTBO-ELM classifier; and 5) inputting test set data into the trained SCMTBO-ELM model to realize fault diagnosis. According to the invention, GIS partial discharge fault diagnosis can be carried out simply and rapidly, the test performance of experimental data is excellent, the fault can be accurately identified by field data, and the method has strong reliability and superiority.

Description

Technical Field

[0001] The invention belongs to the technical field of GIS state monitoring and fault diagnosis, and in particular relates to a GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm. Background Art

[0002] The hazards of GIS partial discharge include the following aspects: 1) insulation material damage; 2) local temperature rise, even leading to thermal breakdown; 3) generation of corrosive gas, accelerating insulation degradation; 4) causing equipment failure, affecting the safety and reliability of the power system. Therefore, it is very important to quickly and accurately identify GIS partial discharge faults. At present, the main methods of fault identification include support vector machine (SVM), extreme learning machine (ELM), various deep learning, neural network, etc. However, the setting of key parameters seriously affects the accuracy of the model in identifying partial discharge.

[0003] At present, there are artificial settings for classifier parameter selection, but it is too dependent on experience, there is contingency, poor scientific basis, and low recognition rate; there are also particle swarm algorithms, genetic algorithms, gray wolf algorithms, etc. to adaptively optimize parameters, but the algorithm converges at a low speed and accuracy, and it is very easy to fall into the situation of local optimal solution, resulting in slow recognition speed and poor accuracy. Therefore, this patent proposes an improved mountaineering team optimization algorithm to automatically optimize the input weight and hidden layer threshold of ELM, and the experiment verifies that it has a very high accuracy rate. Summary of the invention

[0004] In order to solve the above problems, the present invention provides a GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm, which uses refraction reverse learning and Cauchy mutation strategy to improve the mountaineering team optimization algorithm, and automatically optimizes the input weights and hidden layer thresholds of ELM to realize GIS partial discharge fault identification, with rapid detection and high accuracy, thereby improving the reliability of power supply.

[0005] The present invention specifically provides a GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm, and the GIS partial discharge fault identification method comprises the following steps:

[0006] 1) Collect four typical fault signals of partial discharge;

[0007] 2) extracting the mean, root mean square, variance, skewness, kurtosis index, pulse index, peak index, margin index, and waveform index of the partial discharge signal as feature vectors;

[0008] 3) Improved mountaineering team optimization algorithm (SCMTBO) using refraction reverse learning and Cauchy mutation strategy;

[0009] 4) In the training set, the input weights and hidden layer thresholds of the extreme learning machine (ELM) are optimized using the improved mountaineering team optimization algorithm. After reaching the maximum number of iterations, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier;

[0010] 5) Input the test set data into the trained SCMTBO–ELM model to realize fault diagnosis.

[0011] Furthermore, the four typical partial discharge faults in step 1) specifically include tip discharge, particle discharge, air gap discharge, and suspension discharge.

[0012] This paper proposes an optimization algorithm for a mountaineering team, which consists of multiple climbers and an experienced and professional leader, and whose goal is to conquer the top of the mountain, which is the final global solution to the optimization problem.

[0013] In the first phase, the players move towards the leader as follows:

[0014]

[0015] in, is the latest position of the team members, X i is the position of the player in the previous iteration, rand is a random number (0, 1), X leader For the leader position.

[0016] The team members are arranged from the best to the worst fitness, and everyone is guided by the leader and the team member in front of them, so the equation for moving to the top of the mountain is modified to:

[0017]

[0018] Among them, X ii It is the position pointed by the previous player.

[0019] The second stage is the impact of natural disasters.

[0020] Natural disasters such as avalanches may occur during mountaineering, which is equivalent to the local optimal solution X in the algorithm. worst or X Avalanche , the i-th person tries to escape from the local optimal solution X worst or X Avalanche , and through the following heuristic self-help, that is, the individual is freed from being trapped in the local optimum and continues to move towards the optimal solution:

[0021]

[0022] The third stage is collective efforts to respond to disasters.

[0023] When an avalanche occurs, the entire team will work hard to rescue the trapped members, that is, the positions of all members are equal to their average position X Team , the model for this behavior is as follows:

[0024]

[0025] Stage 4: Members may die

[0026] Sometimes, due to the strong avalanche, some members unfortunately die and cannot be saved, that is, the member is removed from the group and replaced by a new member using the following formula:

[0027]

[0028] Among them, X min () is the position of the member with smaller fitness than this member after iteration; X max is the position of the member whose fitness is greater than that of this member after iteration.

[0029] Furthermore, in step 3), the specific steps of improving the mountaineering team optimization algorithm by refraction reverse learning and Cauchy mutation strategy are as follows:

[0030] 3.1 Initializing individual climbers using refraction reverse learning

[0031] The optimal range of the solution on the x-axis is [l,u], the y-axis is the normal line, α and β represent the incident angle and refraction angle respectively, h and h* are the lengths corresponding to the incident and refracted light rays respectively, and O is the midpoint of the optimal range [l,u]. Let the scaling factor k = h / h*, then:

[0032]

[0033] In the formula, x i,j is the position of the i-th player in the population in the j-dimensional position (i=1,2,…,M; j=1,2,…,N), M is the population size, and N is the dimension; For x i,j The refraction reverse position; l j 、u j are the minimum and maximum values ​​of the j-th dimension of the search space, respectively.

[0034] Generate the refraction reverse population according to formula (6)

[0035] 3.2 The Cauchy mutation strategy is used to replace the random member replacement stage of the original mountaineering team algorithm.

[0036] The Cauchy mutation strategy is introduced to improve the ability of the function to find the best fitness globally. The update formula is calculated as follows:

[0037]

[0038] In the formula, is the member position after the Cauchy mutation strategy, cauchy(0,1) is the standard Cauchy distribution function; It means multiplication.

[0039] Furthermore, the principle of the extreme learning machine ELM in step 4) is as follows:

[0040] ELM is a single hidden layer feedforward neural network algorithm, which consists of input layer, hidden layer and output layer. i , t i ), a i =[a i1 ,a i2 ,…a in ] T , is the input of the i-th sample; n is the number of input neurons, is the number of input features; t i =[t i1 ,t i2 ,…,t im ] T , is the model about a i The output of; m is the number of neurons in the output layer, is the number of fault types; sample a i The corresponding output L of the kth hidden layer neuron k (a i )for:

[0041] L k (a i )=G(ω k ·a i +b k )(8)

[0042] In the formula, ω k =[ω k1 ,ω k2 ,...,ω ki ] T is the weight vector between the kth hidden layer neuron and the input layer; b k is the threshold of the kth hidden layer neuron; G is the activation function, using the Sigmoid function; k = 1, 2, ... p, p is the number of hidden layer nodes;

[0043] Sample a i The output t of the corresponding output layer i for:

[0044]

[0045] In the formula, χ kis the weight of the kth hidden layer neuron and the output layer;

[0046] Then the output matrix of q samples is:

[0047] Y=Hχ(10)

[0048] In the formula, Y=[y 1 ,y 2 ,...,y q ] T ; χ=[χ 1 ,x 2 ,... p ];

[0049]

[0050] In the ELM parameter setting, SCMTBO is introduced, and the optimization target is the input layer weight ω k and hidden layer threshold b k ;

[0051] The error rate of ELM is used as the fitness function, and its formula is:

[0052]

[0053] In the formula, r correct is the number of samples correctly classified, r all is the total number of samples;

[0054] The population size and the maximum number of iterations are set. When the maximum number of iterations is reached, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier, and the test set is used for partial discharge fault diagnosis.

[0055] Compared with the prior art, the beneficial effects are:

[0056] (1) The present invention extracts the mean, root mean square, variance, skewness, kurtosis index, pulse index, peak index, margin index, and waveform index of the partial discharge signal as feature vectors, thereby improving the disadvantage of too few feature vectors for partial discharge fault identification.

[0057] (2) The mountaineering team optimization algorithm improved by refractive reverse learning and Cauchy mutation strategy has faster convergence speed and convergence accuracy.

[0058] (3) The SCMTBO-ELM proposed in the present invention has high recognition accuracy and fast operation speed, can accurately diagnose the type of partial discharge fault in GIS, and has strong engineering practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1It is a flow chart of the GIS partial discharge fault identification method of the present invention;

[0060] Figure 2 It is the GIS partial discharge signal diagram;

[0061] Figure 3 It is a schematic diagram of refraction reverse learning;

[0062] Figure 4 It is the original data graph of a certain fault;

[0063] Figure 5 (a)-(d) are the convergence curves of the test function;

[0064] Figure 6 (a)-(d) are the recognition results of different classifiers. DETAILED DESCRIPTION

[0065] The following is a detailed description of a specific implementation of a GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm of the present invention in conjunction with the accompanying drawings.

[0066] like Figure 1 As shown, the GIS partial discharge fault identification method of the present invention comprises the following steps:

[0067] 1) Collect four typical fault signals of partial discharge;

[0068] 2) extracting the mean, root mean square, variance, skewness, kurtosis index, pulse index, peak index, margin index, and waveform index of the partial discharge signal as feature vectors;

[0069] 3) Improved mountaineering team optimization algorithm (SCMTBO) using refraction reverse learning and Cauchy mutation strategy;

[0070] 4) In the training set, the input weights and hidden layer thresholds of the extreme learning machine (ELM) are optimized using the improved mountaineering team optimization algorithm. After reaching the maximum number of iterations, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier;

[0071] 5) Input the test set data into the trained SCMTBO–ELM model to realize fault diagnosis.

[0072] like Figure 2 As shown, the four typical partial discharge faults in step 1) specifically include tip discharge, particle discharge, air gap discharge, and suspension discharge.

[0073] This paper proposes an optimization algorithm for a mountaineering team, which consists of multiple climbers and an experienced and professional leader, and whose goal is to conquer the top of the mountain, which is the final global solution to the optimization problem.

[0074] In the first phase, the players move towards the leader as follows:

[0075]

[0076] in, is the latest position of the team members, X i is the position of the player in the previous iteration, rand is a random number (0, 1), X leader r is the leader position.

[0077] The team members are arranged from the best to the worst fitness, and everyone is guided by the leader and the team member in front of them, so the equation for moving to the top of the mountain is modified to:

[0078]

[0079] Among them, X ii It is the position pointed by the previous player.

[0080] The second stage is the impact of natural disasters.

[0081] Natural disasters such as avalanches may occur during mountaineering, which is equivalent to the local optimal solution X in the algorithm. worst or X Avalanche , the i-th person tries to escape from the local optimal solution X worst or X Avalanche , and through the following heuristic self-help, that is, the individual is freed from being trapped in the local optimum and continues to move towards the optimal solution:

[0082]

[0083] The third stage is collective efforts to respond to disasters.

[0084] When an avalanche occurs, the entire team will work hard to rescue the trapped members, that is, the positions of all members are equal to their average position XTeam. The model of this behavior is as follows:

[0085]

[0086] Stage 4: Members may die

[0087] Sometimes, due to the strong avalanche, some members unfortunately die and cannot be saved, that is, the member is removed from the group and replaced by a new member using the following formula:

[0088]

[0089] Among them, X min () is the position of the member with smaller fitness than this member after iteration; Xmax is the position of the member whose fitness is greater than that of this member after iteration.

[0090] In step 3), the specific steps of refraction reverse learning and Cauchy mutation strategy to improve the mountaineering team optimization algorithm are as follows:

[0091] 3.1 Initializing individual climbers using refraction reverse learning

[0092] like Figure 3 Shown is the schematic diagram of refraction reverse learning.

[0093] The optimal solution range on the x-axis is [l,u], the y-axis is the normal line, α and β represent the incident angle and refraction angle respectively, h and h* are the lengths corresponding to the incident and refracted light rays respectively, and O is the midpoint of the optimal solution range [l,u]. Let the scaling factor k = h / h*, then:

[0094]

[0095] In the formula, x i,j is the position of the i-th player in the population in the j-dimensional position (i=1,2,…,M; j=1,2,…,N), M is the population size, and N is the dimension; For x i,j The refraction reverse position; l j 、u j are the minimum and maximum values ​​of the j-th dimension of the search space, respectively.

[0096] Generate the refraction reverse population according to formula (6)

[0097] 3.2 The Cauchy mutation strategy is used to replace the random member replacement stage of the original mountaineering team algorithm.

[0098] The Cauchy mutation strategy is introduced to improve the ability of the function to find the best fitness globally. The update formula is calculated as follows:

[0099]

[0100] In the formula, is the position of the member after the Cauchy mutation strategy, cauchy(0,1) is the standard Cauchy distribution function; ⊕ means multiplication. The specific steps of improving the mountaineering team optimization algorithm by combining refraction reverse learning and Cauchy mutation strategy are as follows: Figure 4 shown.

[0101] The principle of the extreme learning machine ELM in step 4) is as follows:

[0102] ELM is a single hidden layer feedforward neural network algorithm, which consists of input layer, hidden layer and output layer. i , t i), a i =[a i1 ,a i2 ,…a in ] T , is the input of the i-th sample; n is the number of input neurons, is the number of input features; t i =[t i1 ,t i2 ,…,t im ] T , is the model about a i The output of; m is the number of neurons in the output layer, is the number of fault types; sample a i The corresponding output L of the kth hidden layer neuron k (a i )for:

[0103] L k (a i )=G(ω k ·a i +b k ) (8)

[0104] In the formula, ω k =[ω k1 ,ω k2 ,...,ω ki ] T is the weight vector between the kth hidden layer neuron and the input layer; b k is the threshold of the kth hidden layer neuron; G is the activation function, using the Sigmoid function; k = 1, 2, ... p, p is the number of hidden layer nodes;

[0105] The output ti of the output layer corresponding to sample ai is:

[0106]

[0107] In the formula, χ k is the weight of the kth hidden layer neuron and the output layer;

[0108] Then the output matrix of q samples is:

[0109] Y=Hχ (10)

[0110] In the formula, Y=[y 1 ,y 2 ,...,y q ] T ; χ=[χ 1 ,x 2 ,... p ];

[0111]

[0112] In the ELM parameter setting, SCMTBO is introduced, and the optimization target is the input layer weight ω k and hidden layer threshold b k ;

[0113] The error rate of ELM is used as the fitness function, and its formula is:

[0114]

[0115] In the formula, r correct is the number of samples correctly classified, r a ll is the total number of samples;

[0116] The population size and the maximum number of iterations are set. When the maximum number of iterations is reached, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier, and the test set is used for partial discharge fault diagnosis classification.

[0117] In order to verify the superiority of SCMTBO, MTBO, PSO and GWO algorithms are used for comparison. The population size is set to 100, the number of iterations is set to 1000, and four groups of test functions are used to verify their convergence speed and accuracy. The test functions are shown in Table 1. Figure 5 (a) shows the convergence curve of the test function F1, as shown in Figure 5 (b) shows the convergence curve of the test function F2, as shown in Figure 5 (c) shows the convergence curve of the test function F3. Figure 5 (d) shows the convergence curve diagram of the test function F4.

[0118] Table 1

[0119]

[0120]

[0121] Depend on Figure 5 It can be seen that the improved mountaineering team optimization algorithm proposed in this patent has an absolute advantage in terms of convergence speed and accuracy, reflecting its superiority.

[0122] The experiment collected four typical defects: tip discharge, particle discharge, air gap discharge, and suspension discharge for verification. 120 sets of samples were collected for each defect, a total of 480 sets of samples, 400 sets for training sets, and 80 sets for test sets. The signals are as follows Figure 2The mean, root mean square, variance, skewness, kurtosis index, pulse index, peak index, margin index, and waveform index of the signal are extracted as feature vectors and input into SCMTBO-ELM for training and testing. The diagnostic results are shown in Figure 6 As shown in (a)-(d), the accuracy rates of the four classifiers are 98.75%, 90%, 95%, and 88.75%, respectively, and the time taken is 0.856s, 1.726s, 1.524s, and 1.591s, respectively. The SCMTBO-ELM classifier has obvious advantages. It can be seen that the method proposed in the present invention improves the accuracy of GIS partial discharge fault diagnosis and shortens the operation time.

[0123] It can be seen from the above experiments that the present invention can accurately judge the partial discharge fault of GIS equipment.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention and are not intended to limit the present invention. A person skilled in the art should understand that the specific implementation of the present invention can be modified or replaced by equivalents, but these modifications or changes are within the scope of protection of the pending claims.

Claims

1. A GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm, characterized in that: The GIS partial discharge fault identification method comprises the following steps: 1) Collect four typical fault signals of partial discharge; 2) extracting the mean, root mean square, variance, skewness, kurtosis index, pulse index, peak index, margin index, and waveform index of the partial discharge signal as feature vectors; 3) Improve the mountaineering team optimization algorithm SCMTBO using refraction reverse learning and Cauchy mutation strategy; 4) In the training set, the input weights and hidden layer thresholds of the extreme learning machine ELM are optimized using the improved mountaineering team optimization algorithm. After reaching the maximum number of iterations, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier; 5) Input the test set data into the trained SCMTBO–ELM model to realize fault diagnosis.

2. A GIS partial discharge fault identification method based on an improved mountaineering team optimization algorithm according to claim 1, characterized in that: The four typical partial discharge faults in step 1) include tip discharge, particle discharge, air gap discharge, and suspension discharge.

3. The GIS partial discharge fault identification method based on the improved mountaineering team optimization algorithm according to claim 1 is characterized in that: In step 3), the specific steps of refraction reverse learning and Cauchy mutation strategy to improve the mountaineering team optimization algorithm SCMTBO are as follows: 3.1 Use refraction reverse learning to initialize individual climbers; 3.2 The Cauchy mutation strategy is used to replace the random member replacement stage of the original mountaineering team algorithm.

4. The GIS partial discharge fault identification method based on the improved mountaineering team optimization algorithm according to claim 3 is characterized in that: Step 3.1 The specific method of using refraction reverse learning to initialize individual climbers is: The optimal range of the solution on the x-axis is [l,u], the y-axis is the normal line, α and β represent the incident angle and the refraction angle respectively, h and h* are the lengths corresponding to the incident and refracted light rays respectively, and O is the midpoint of the optimal range [l,u]; let the scaling factor k = h / h*, then: In the formula, x i,j is the position of the i-th player in the population in the j-dimensional position (i=1,2,…,M; j=1,2,…,N), M is the population size, and N is the dimension; For x i,j The refraction reverse position; l j 、u j are the minimum and maximum values ​​of the j-th dimension of the search space respectively; Generate the refraction reverse population according to the above formula 5. The GIS partial discharge fault identification method based on the improved mountaineering team optimization algorithm according to claim 4 is characterized in that: In step 3.2, the Cauchy mutation strategy is used to replace the random member replacement stage of the original mountaineering team algorithm, specifically: The Cauchy mutation strategy is introduced to improve the ability of the function to find the best fitness globally; the update formula is calculated as follows: In the formula, is the member position after the Cauchy mutation strategy, cauchy(0,1) is the standard Cauchy distribution function; ⊕ means multiplication.

6. The GIS partial discharge fault identification method based on the improved mountaineering team optimization algorithm according to claim 1 is characterized in that: The principle of the extreme learning machine ELM in step 4) is as follows: ELM is a single hidden layer feedforward neural network algorithm, which consists of input layer, hidden layer and output layer. i , t i ), a i =[a i1 ,a i2 ,…a in ] T , is the input of the i-th sample; n is the number of input neurons, and is the number of input features; t i =[t i1 ,t i2 ,…,t im ] T , is the model about a i The output of; m is the number of neurons in the output layer, is the number of fault types; sample a i The corresponding output L of the kth hidden layer neuron k (a i )for: L k (a i )=G(ω k ·a i +b k ) In the formula, ω k =[ω k1 ,ω k2 ,...,ω ki ] T is the weight vector between the kth hidden layer neuron and the input layer; b k is the threshold of the kth hidden layer neuron; G is the activation function, using the Sigmoid function; k = 1, 2, ... p, p is the number of hidden layer nodes; Sample a i The output t of the corresponding output layer i for: In the formula, χ k is the weight of the kth hidden layer neuron and the output layer; Then the output matrix of q samples is: where Y = [y1, y2,..., y q T ; χ = [χ1, χ2,... χ p ;​ In the ELM parameter setting, SCMTBO is introduced, and the optimization target is the input layer weight ω k and hidden layer threshold b k ; The error rate of ELM is used as the fitness function, and its formula is: In the formula, r correct is the number of samples correctly classified, r all is the total number of samples; The population size and the maximum number of iterations are set. When the maximum number of iterations is reached, the coordinates with the optimal solution are input into the ELM model as the optimal parameter combination to establish the SCMTBO-ELM classifier, and the test set is used for partial discharge fault diagnosis classification.

Images