GIS partial discharge mode identification method, system and device and storage medium

Abstract

The invention relates to a GIS partial discharge mode identification method, system and device, and a storage medium. The method comprises the following steps: selecting on-site multi-platform GIS partial discharge monitoring data to form a training sample set; performing model parameter optimization on a pre-established DBN network model by using an improved PSO algorithm, and training based on the training sample set data to obtain an optimal DBN network model; and carrying out partial discharge mode identification on the target GIS switch equipment based on the obtained optimal DBN network model to obtain a partial discharge mode identification result of the target GIS switch equipment. The method can be widely applied to the sub-field of electrical equipment monitoring and fault diagnosis in the field of power systems.

Description

Technical Field

[0001] The present invention belongs to the sub-field of electrical equipment monitoring and fault diagnosis in the field of power systems, and relates to GIS partial discharge monitoring and identification technology, and specifically to a GIS partial discharge pattern recognition method, system, device and storage medium based on a particle swarm optimization deep belief network. Background Art

[0002] Gas insulated metal-enclosed switchgear (GIS) is an integrated device that integrates key switch components in substations. It has the advantages of compact structure, small footprint, less electromagnetic pollution, strong anti-interference ability, less maintenance workload, safety and reliability, and high stability. It is widely used in high-voltage substations. However, with the long-term operation of GIS equipment, it is inevitable that insulation defects will induce partial discharge due to various reasons. In addition, GIS equipment is an integrated structure, which poses challenges such as large workload in fault source finding and difficulty in formulating maintenance plans. Therefore, accurate identification of partial discharge types plays an important role in fault judgment and maintenance plan formulation of GIS equipment.

[0003] At present, many scholars at home and abroad have conducted a lot of research on partial discharge pattern recognition algorithms. Among them, the more commonly used recognition algorithms include back propagation neural network (BP), support vector machine (SVM) and convolutional neural network (CNN). These commonly used methods have certain recognition capabilities for partial discharge, but they also have certain limitations. BP network is prone to falling into the minimum value; SVM needs to set a reasonable function and penalty coefficient, which is sensitive to the distribution of data and may increase the amount of calculation when there is a lot of data; the architecture of CNN lacks consideration of the importance of different features, which may lead to an increase in the amount of calculation.

[0004] In this regard, the deep belief network (DBN) uses unsupervised pre-training and supervised fine-tuning to process data, integrating the characteristics of deep learning and feature learning. It has strong data analysis capabilities and can solve the problems of traditional commonly used partial discharge recognition algorithms. In recent years, this algorithm has been widely used in image recognition, partial discharge recognition, and load forecasting. "Kong Xiangyu, Zheng Feng, E Zhijun, et al. Short-term load forecasting method based on deep belief network [J]. Automation of Electric Power Systems, 2018, 42(05): 133-139." The short-term load forecasting method was carried out using the deep belief network method, and it was found that this method has higher feature learning ability. "Zhang Xinbo, Tang Ju, Pan Cheng, et al. Deep belief network method for partial discharge pattern recognition [J]. Power System Technology, 2016, 40(10): 3272-3278." The extracted PRPD map features were trained and classified using the deep belief network method, and good recognition results were achieved. However, there is a problem that the artificial weight parameter setting has a great influence on the accurate recognition rate of partial discharge. Summary of the invention

[0005] In view of the above problems, the purpose of the present invention is to provide a GIS local discharge pattern recognition method, system, device and storage medium, which utilizes particle swarm (PSO) to optimize the local discharge pattern recognition of deep belief network (DBN), and has a higher accurate recognition rate than traditional DBN network, multi-layer feedforward neural network (BP), support vector machine (SVM) and convolutional neural network (CNN). The method optimizes the weight parameters of the DBN network through the PSO algorithm, reduces the influence of the DBN network weight parameters set subjectively on the local discharge pattern recognition, and improves the network's learning ability for local discharge characteristics.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a GIS partial discharge pattern recognition method, comprising the following steps:

[0008] The on-site multi-platform GIS partial discharge monitoring data are selected to form a training sample set;

[0009] The improved PSO algorithm is used to optimize the model parameters of the pre-established DBN network model, and the optimal DBN network model is obtained based on the training sample set data.

[0010] Based on the obtained optimal DBN network model, partial discharge pattern recognition is performed on the target GIS switch equipment to obtain the partial discharge pattern recognition result of the target GIS switch equipment.

[0011] Furthermore, the selection of on-site multi-platform GIS partial discharge monitoring data to form a training sample set includes:

[0012] Obtain partial discharge monitoring data of multi-platform GIS switchgear on site;

[0013] Preprocess and normalize the acquired partial discharge monitoring data;

[0014] The normalized partial discharge monitoring data is divided into a training set and a test set according to a preset ratio as a training sample set.

[0015] Furthermore, the partial discharge monitoring data includes: skewness, correlation coefficient, discharge amount factor, asymmetry, steepness and number of local peaks.

[0016] Furthermore, the improved PSO algorithm is used to optimize the model parameters of the pre-established DBN network model, and the optimal DBN network model is obtained based on the training sample set data, including:

[0017] The improved PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration times in the DBN network model to obtain the optimal particle vector.

[0018] The optimal particle vector is set as the corresponding DBN network model parameter, the initial optimal DBN network model is established and trained using the training set data;

[0019] The optimal DBN network model obtained through training is used to identify, classify and verify the test set data to obtain the optimal DBN network model that meets the preset accuracy.

[0020] Furthermore, the improved PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration number parameters in the DBN network model to obtain the optimal particle vector, including:

[0021] Establish DBN network model;

[0022] The optimized PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration times of the DBN network model to obtain the optimal particle vector.

[0023] Furthermore, the DBN network model includes:

[0024] Input layer, first to third RBM layers, BP layer and output layer;

[0025] The input layer is used to obtain input feature data;

[0026] The first to third RBM layers are used to extract feature information of input partial discharge data and supervised reverse fine-tuning model parameters through forward unsupervised autonomous learning;

[0027] The BP layer is used to reduce the error between the final recognition result and the actual result;

[0028] The output layer is used to output annotated partial discharge data.

[0029] Furthermore, the optimized PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration number parameters of the DBN network model to obtain the optimal particle vector, including:

[0030] ①Determine the fitness function of the particle swarm;

[0031] ② Determine the relevant parameter range of the particle swarm and initialize the correlation coefficient of the particles;

[0032] ③ Map the initialized DBN network model parameters to the dimensions of the particles;

[0033] ④ Iterate and optimize the weight parameters of the DBN network model according to the current parameters of each particle, and calculate the fitness value of each particle after the kth iteration of value search;

[0034] ⑤ Update the optimal value of each particle and population in the current iteration round according to the fitness value and extreme value before and after the particle optimization, and update the inertia coefficient and learning factor of the particle according to the current number of iterations k;

[0035] ⑥ Use the updated particle coefficient to update the speed and position of all particles in the population, and determine whether the current iteration number in the particle optimization process is equal to the maximum iteration number of the particle set in advance: If the current iteration number of the particle k≠T, return to step ⑤ to continue the particle iterative optimization process; if the current iteration number k=T, the PSO algorithm iterative optimization ends and the optimal particle vector is obtained.

[0036] In a second aspect, the present invention provides a GIS partial discharge pattern recognition system, comprising:

[0037] The data acquisition module is used to select the on-site multi-platform GIS partial discharge monitoring data to form a training sample set;

[0038] The model training module is used to optimize the model parameters of the pre-established DBN network model using the improved PSO algorithm, and obtain the optimal DBN network model based on the training sample set data;

[0039] The partial discharge recognition module is used to perform partial discharge pattern recognition on the target GIS switch device based on the obtained optimal DBN network model to obtain the partial discharge pattern recognition result of the target GIS switch device.

[0040] In a third aspect, the present invention provides a computer-readable storage medium storing one or more programs, wherein the one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any method.

[0041] In a fourth aspect, the present invention provides a computing device, comprising: one or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, and the one or more programs include instructions for executing any method.

[0042] The present invention adopts the above technical solution, which has the following advantages:

[0043] 1. The present invention uses the PSO algorithm to optimize the parameters of the DBN network model. The PSO algorithm can enable the DBN network to obtain better network parameters, further enhance the ability of the deep belief network to learn and mine the potential relationship of data, and obtain better partial discharge identification effect. Compared with BP, CNN, SVM, and traditional DBN networks, this network model has better identification effects on four typical partial discharge types.

[0044] 2. Traditional recognition models such as BP, SVM and CNN are difficult to identify partial discharge types such as suspended, free metal particles and solid insulating gas with certain similar characteristics in the PRPD spectrum, while the traditional DBN network model is able to perform high-precision identification, and the recognition effect of the DBN network model optimized by the particle swarm algorithm will be better. The largest improvement in accuracy is the recognition of free metal particle discharge, and the recognition of tip discharge and free metal particle partial discharge has a higher degree of credibility.

[0045] 3. PSO is used to optimize the model parameters of the DBN network. Compared with multiple manual parameter adjustments, it can effectively reduce the impact of the parameters on the final accurate recognition rate of the algorithm. Compared with the traditional DBN network, it has improvements in the recognition of suspended discharge and free metal particle discharge. The final PSO-DBN network has a higher accurate recognition rate for both single partial discharge type recognition and overall partial discharge recognition.

[0046] Therefore, the present invention can be widely applied to the sub-fields of electrical equipment monitoring and fault diagnosis in the field of power systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Throughout the accompanying drawings, the same reference numerals are used to represent the same components. In the accompanying drawings:

[0048] Figure 1 is a flow chart of a PSO optimized DBN network provided by an embodiment of the present invention;

[0049] Figure 2a to Figure 2b is a typical insulation defect partial discharge PRPD spectrum provided by an embodiment of the present invention, wherein: Figure 2a For tip discharge, Figure 2b For free metal particles, Figure 2c is the suspension potential, Figure 2d It is solid insulation air gap discharge;

[0050] Figure 3 is a structural diagram of a DBN model provided by an embodiment of the present invention;

[0051] Figure 4 The time series prediction of transformer operation trend provided by the embodiment of the present invention;

[0052] Figure 5 It is a GIS equipment model diagram provided by an embodiment of the present invention;

[0053] Figure 6 is a particle swarm fitness curve provided by an embodiment of the present invention;

[0054] Figure 7 It is the accuracy recognition rate of different partial discharge types by different models provided by the embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiment of the present invention clearer, the technical solution of the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings of the embodiment of the present invention. Obviously, the described embodiment is a part of the embodiment of the present invention, not all of the embodiments. Based on the described embodiment of the present invention, all other embodiments obtained by ordinary technicians in this field belong to the scope of protection of the present invention.

[0056] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0057] In order to reduce the influence of DBN network weight parameters set by humans on partial discharge pattern recognition, give full play to the advantages of DBN network deep learning, and further improve the accuracy of partial discharge recognition of GIS equipment. In some embodiments of the present invention, a GIS partial discharge pattern recognition method based on particle swarm optimization (PSO) to optimize DBN network is provided. First, the on-site multi-platform GIS partial discharge monitoring data is selected to form a sample set; secondly, the improved PSO algorithm is used to optimize the parameters such as network structure, learning rate and forward and reverse iteration times in the DBN model in combination with the sample set data, and the parameters obtained by PSO iteration are set as the corresponding DBN network weight parameters to establish the optimal DBN network model; finally, the model is used to identify, classify and verify the partial discharge sample set of the shore power offshore power platform of Bohai Oilfield.

[0058] Correspondingly, in other embodiments of the present invention, a GIS local discharge pattern recognition system, device and storage medium are provided.

[0059] Example 1

[0060] like Figure 1 As shown, this embodiment provides a GIS partial discharge pattern recognition method, and its overall process is divided into three parts: data processing, PSO algorithm optimization of DBN network model parameters, and partial discharge recognition. Specifically, it includes the following steps:

[0061] 1) Select on-site multi-platform GIS partial discharge monitoring data to form a training sample set;

[0062] 2) Use the improved PSO algorithm to optimize the model parameters of the pre-established DBN network model, and obtain the optimal DBN network model based on the training sample set data;

[0063] 3) Based on the obtained optimal DBN network model, the local discharge mode of the target GIS switch device is identified to obtain the local discharge mode recognition result of the target GIS switch device.

[0064] Furthermore, the above step 1) includes the following steps:

[0065] 1.1) Obtain partial discharge monitoring data of multi-platform GIS switchgear on site;

[0066] 1.2) Preprocess and normalize the acquired partial discharge monitoring data to ensure that the original data is relatively undistorted and easy to analyze and calculate;

[0067] Among them, the normalization formula is:

[0068]

[0069] In the formula, x represents the size of a variable in the data set, x' represents the size of the variable x in the data set after normalization, max(x) and min(x) are the maximum and minimum values ​​of each type of data in the partial discharge data set, respectively.

[0070] 1.3) Divide the normalized partial discharge monitoring data into a training set and a test set according to a preset ratio as a training sample set.

[0071] Furthermore, in the above step 1.1), the partial discharge signal of the GIS switchgear can be detected by single or multiple methods such as light, electricity and sound, among which the ultra-high frequency detection method is more commonly used. In the ultra-high frequency detection method, the partial discharge phase distribution diagram (phase resolved partial discharge, PRPD) is often used to analyze and extract the characteristics of different discharge types. The PRPD spectrum mainly shows the relationship between the phase (180-360°) and the discharge amplitude and the number of discharges n of the partial discharge, which can be regarded as another pulse sequence distribution spectrum (phase resolved pulse sequence, PRPS) for partial discharge analysis in a certain period. Superposition or projection.

[0072] In actual engineering, the PRPD diagrams of four typical partial discharge fault types commonly seen in GIS switchgear are as follows: Figure 2a to Figure 2d As shown in the figure, it can be seen that the tip discharge is mainly concentrated near 270°, while the free metal particles have less or almost no discharge near 90° and 270°, and mainly discharge more at other phases; the suspension discharge is mainly distributed in 0-90° and 180-270°, and shows a certain suspension characteristic in the spectrum; the solid insulation air gap discharge mainly occurs at the rising edge of the corresponding positive and negative half-cycles of the power frequency voltage, which is similar to the distribution phase of the suspension discharge, but the amplitude of the solid insulation air gap discharge is more dispersed.

[0073] For the identification of GIS partial discharge types, obtaining effective features of different fault types is the key point for classification. The correlation between the extracted features of the spectrum and partial discharge is also one of the important reasons affecting the accuracy of classification. When extracting partial discharge type features for identifying partial discharge patterns, the time domain analysis method will reduce the accuracy of feature extraction through transformation results due to the increase in the degree of attenuation distortion during signal transmission, and the field application effect is poor; the effect of using statistical parameters of statistical analysis method to characterize different types of local faults is better than that of time domain analysis method. This method is also the most widely used method for partial discharge identification of electrical equipment. The commonly used statistical operators for extracting and characterizing partial discharge features based on the shape features of PRPD spectrum are: skewness Sk , correlation coefficient Cc, discharge factor Q, asymmetry ψ, steepness k u And the number of local peaks P e wait.

[0074] S k The calculation formula is:

[0075]

[0076] Where w is the number of phase windows contained in the half-period class of the spectrum, x i is the phase corresponding to the data point in the i-th phase window in the spectrum, p i , μ, σ 2 are the probability, mean and variance of the data point phase in the i-th phase window respectively.

[0077] The calculation formula of Cc is:

[0078]

[0079] In the formula, The positive and negative signs represent the average discharge amount within the i-th phase window in the positive and negative half-cycles of the plotted spectrum, respectively.

[0080] The calculation formula of Q is:

[0081]

[0082] In the formula, and Respectively represent the discharge amount and The number of discharges, where positive and negative represent the positive and negative half cycles of the spectrum.

[0083] The calculation formula of ψ is:

[0084]

[0085] In the formula, and They respectively represent the starting phase angles of discharge in the positive and negative cycles in the spectrum.

[0086] k u The calculation formula is:

[0087]

[0088] In the formula, the definition of each parameter is the same as that of skewness, and the calculation method of the three parameters adopts the corresponding parameter calculation formula in skewness.

[0089] In the above characteristic operator S k , P e , ku It mainly characterizes the difference in the shape of the partial discharge spectrum; Cc, Q, ψ mainly characterize the difference in the contour of the positive and negative half-cycles of the spectrum. This embodiment will use the PSO-DBN model to learn and test these features and the corresponding partial discharge fault types.

[0090] Further, the above step 2) includes the following steps:

[0091] 2.1) Use the improved PSO algorithm to optimize the network structure, learning rate, and forward and reverse iteration times in the DBN network model to obtain the optimal particle vector;

[0092] 2.2) Setting the optimal particle vector to the corresponding DBN network model parameter, establishing the initial optimal DBN network model and training it using the training set data;

[0093] 2.3) Use the optimal DBN network model obtained through training to identify, classify and verify the test set data to obtain the optimal DBN network model that meets the preset accuracy.

[0094] Further, the above step 2.1) includes the following steps:

[0095] 2.1.1) Establish DBN network model;

[0096] 2.1.2) Use the optimized PSO algorithm to optimize the relevant parameters of the DBN network model and obtain the optimal particle vector.

[0097] Further, in the above step 2.1.1), if Figure 3 As shown, the DBN network model established in this embodiment adopts a 3-layer RBM structure, which includes an input layer, the first to third RBM layers, a BP layer and an output layer. Among them, the input layer is used to obtain input feature data; the first to third RBM layers have the following functions: forward unsupervised autonomous learning to extract the feature information of the input partial discharge data and supervised reverse fine-tuning model parameters. Each layer of the RBM model of the network will perform unsupervised deep learning of partial discharge data features, and use the results of this layer as the input information of the previous layer to conduct in-depth learning layer by layer to avoid the network falling into the local optimum; the BP layer is used to reduce the error between the final recognition result and the actual result, improve the data recognition rate, fine-tune the model parameters to the lower layer, further optimize the global parameters, reduce the error of the final output result of the model, and improve the recognition accuracy of the partial discharge type; the output layer is used to output the labeled partial discharge data.

[0098] Furthermore, in the above step 2.1.2), in this embodiment, the number of neurons, learning factor, fine-tuning factor and number of forward and reverse learning iterations in the DBN network model are iteratively optimized by the PSO optimization algorithm. In the PSO optimization algorithm, the dimension of each particle is set to 7 dimensions, that is, each particle contains the number of neurons a of each layer of RBM in the DBN network model. i (i=1,2,3), unsupervised learning rate factor η1, supervised fine-tuning factor η2, forward iteration training times u1 and reverse fine-tuning times u2, that is, the particle vector is X=(a1,a2,a3,η1,η2,u1,u2). After the iteration of the particle swarm algorithm, the elements in the obtained optimal particle vector are set as parameters in the DBN network model to achieve the goal of optimizing the DBN network.

[0099] Specifically, Figure 4 As shown, the particle swarm algorithm includes the following steps:

[0100] ①Determine the fitness function of the particle swarm.

[0101] In this embodiment, the error rate of training data classification and recognition is used as the optimization objective function value (fitness), and whether to update the particle and whether to record the individual extreme value and group extreme value at this time is determined by whether the fitness is reduced. The fitness function is expressed as:

[0102]

[0103] In the formula, F ij is the fitness of the jth particle in the i-th iteration; n is the number of test samples, p l and t l are the predicted category and the actual category, respectively.

[0104] ② Determine the relevant parameter range of the particle swarm and initialize the correlation coefficient of the particles.

[0105] Specifically, determining the relevant parameter range of the particle swarm mainly refers to determining the speed v max and v min , position range, inertia coefficient w max and w min , learning factor c max and c min , the size N of the population and the maximum number of iterations T of the population, and randomly initialize the correlation coefficient of the particles according to the determined range. In this embodiment, the number of particles is set to 25, and the “number of particles” refers to the size N of the population.

[0106] ③ Map the initialized DBN network model parameters to the dimensions of the particles.

[0107] ④ According to the current parameters of each particle, the weight parameters of the DBN network model are iteratively optimized, and the fitness value of each particle is calculated after the kth iteration of the value search is completed.

[0108] ⑤ Update the optimal value of each particle and population in the current iteration round according to the fitness value and extreme value before and after the particle optimization. At the same time, update the particle's inertia coefficient and learning factor according to the current iteration number k (update with weights and learning factors that change with the increase in iteration number). The updated calculation formulas are:

[0109]

[0110] Specifically, whether to update the optimal value of each particle in the current iteration round and the optimal value of the population is determined based on whether the fitness of the particle decreases before and after optimization and whether the extreme value of other particles is better than that of itself. In each iteration, the particle calculates the current position fitness value and compares it with the previously recorded individual optimal fitness value. If the current fitness value is better, the individual optimal position is updated to the current position. The optimal value of the population is updated by selecting the information of the best particle from all particles.

[0111] ⑥ Use the updated particle coefficients (including "inertia coefficient" and "learning factor") to update the speed and position of all particles in the population, and determine whether the current number of iterations in the particle optimization process is equal to the maximum number of particle iterations set in advance: If the current number of iterations of the particle k≠T, return to step ⑤ to continue the particle iterative optimization process; if the current number of iterations k=T, the PSO algorithm iterative optimization ends and the optimal particle vector is obtained.

[0112] Furthermore, in the above step 2.2), the optimal particle vector is set to the corresponding DBN network model parameters, that is, the values ​​of each dimension in the optimal particle vector are set to the corresponding DBN network model parameters, that is, the number of neurons in each layer, the learning factor and the number of fine-tuning times, to construct the initial optimal DBN network model.

[0113] Example 2

[0114] This embodiment takes the GIS switchgear model 1HZC2320054-02 / 220kV in the shore power offshore power platform of Bohai Oilfield as an example. The platform adopts the UHF detection method to perform online monitoring of the equipment, and describes the specific implementation mode of the present invention in detail.

[0115] In order to verify the accuracy of the partial discharge fault pattern recognition model adopted in the present invention, monitoring historical data of the same model as the equipment were selected from the historical fault database of the shore power offshore power platform equipment group of Bohai Oilfield, and a total of 352 partial discharge fault data were combined, including 88 pieces of each defect data. Among them, 240 fault data (60 pieces of each defect) were selected as the training set, and the remaining 112 fault data (28 pieces of each defect) were selected as the test set for model training and recognition.

[0116] The particle-related parameters of the optimization algorithm PSO in the DBN network are set as follows: the number of neurons in the three layers ranges from 5 to 120, 10 to 100, and 20 to 80 respectively; the learning rates η1 and η2 range from 0.01 to 0.1 and 0.001 to 0.1 respectively; the forward and reverse fine-tuning times u1 and u2 range from 300 to 1000; the inertia coefficient w max =1.2, w min =0.4; learning coefficient c max =2, c min = 0.4; maximum number of iterations T = 40; number of particles q = 25; speed v max =2, v min =-2; the activation function of the DBN network adopts the Sigm function; the number of training samples in each forward and reverse direction is 60 and 30 respectively.

[0117] The combined data set selected by the present invention is a balanced data set, so the accuracy ACC of the recognition result of each algorithm on the test data set is used as the evaluation standard of the corresponding algorithm, and its evaluation expression is:

[0118]

[0119] In the formula, m is the total number of samples in the test set, T p is the partial discharge identification prediction result of the adopted algorithm model, T f Output results for actual PD of the test set.

[0120] Before testing and training the network, the weight parameters of the DBN network are iteratively optimized. The fitness value of the particle reaches the minimum when the number of iterations is close to 15, and the particle vector element has reached the optimal value.

[0121] The optimized DBN network algorithm is trained and tested on the data set mentioned above. The recognition results of the test set are shown in the form of a confusion matrix as shown in Table 1. In the table (the numbers in the matrix represent the number of partial discharges), the ordinate represents the actual partial discharge output type in the data set, and the abscissa represents the corresponding partial discharge output type after prediction using the optimized deep belief network algorithm model. To facilitate the display of the confusion matrix, the partial discharge type is simplified in the confusion matrix shown in Table 1, that is, the letters J, Q, X, and K are used to represent: tip discharge, solid insulation air gap discharge, suspension discharge, and free metal particle discharge.

[0122] Table 1 PSO-DBN model recognition confusion matrix

[0123]

[0124] In order to verify the accurate recognition rate of the PSO-DBN model, this embodiment selected four other commonly used algorithms for recognition comparison, including BP, CNN, SVM and DBN network without PSO optimization. The relevant configuration parameters of these four algorithms are as follows:

[0125] (1) In the BP network, the target training error is set to 1e-6, the maximum number of iterations is set to 1000, and the learning rate is set to 0.01.

[0126] (2) In the SVM recognition algorithm, the commonly used radial basis function (RBF) is selected as the kernel function of the recognition algorithm. This kernel function is simpler to set than other kernel functions. The parameter of the kernel function is set to 0.01 and its penalty factor is set to 10.

[0127] (3) In the CNN network, the maximum number of training times is set to 1000; two convolutional layers and a pooling layer are set, where the number of convolution kernels is 16 and 32 respectively, and the step sizes of convolution and pooling are both 1. The Rectified Linear Unit function (Relu function for short) is used as the activation function of the network.

[0128] (4) The non-optimized DBN network adopts a three-layer RBM structure. The hidden layer neurons of the three RBM layers are set to 100, 80, and 60 respectively; the number of iterations of each RBM layer is set to 500; the forward and reverse fine-tuning learning rates are set to 0.04 and 0.01 respectively; the number of reverse fine-tuning is set to 600; the number of forward and reverse training samples per time is still set to 60 and 30 respectively; the output layer is a softmax layer, and its corresponding partial discharge type classification nodes are set to 4.

[0129] The four algorithms mentioned above were also trained and tested on the original data set, and the overall PD type recognition results of the test set of each algorithm are shown in Table 2. According to the recognition results of the four typical PD types shown in Table 2, the BP network algorithm has the lowest overall recognition rate of the test set among the four non-combined algorithms, which is only 87.5%, while the deep belief network has the highest overall recognition rate, which is 96.43%. On the basis of this network model, the overall recognition rate after adding the improved particle swarm algorithm to optimize its structural parameters is higher, reaching 98.21%. The overall recognition rates of the other two algorithms are higher than those of the BP network, from low to high, SVM: 91.07%, CNN: 93.75%.

[0130] Table 2 Recognition results of different models

[0131] Algorithm Model Test set accuracy (ACC) / % BP 87.50％ SVM 91.07％ CNN 93.75％ DBN 96.43％ PSO-DBN 98.21％

[0132] The corresponding output confusion matrix results of the remaining four algorithm models after testing are shown in Table 3. To facilitate the display of the confusion matrix, the simplified letters in Table 1 are used to represent different partial discharge types.

[0133] Table 3 Recognition confusion matrix of the same model

[0134]

[0135]

[0136] As can be seen from Tables 2 and 3, the BP network has a relatively low recognition rate for K-type partial discharges. In the results, 6 data were misidentified as X-type partial discharges, with an accuracy rate of only 78.57%. Compared with the BP network, the SVM has a 3.57% higher recognition rate for K-type partial discharges, and the recognition rates for Q-type partial discharges and X-type partial discharges are even higher than those for K-type partial discharges, with an improvement of 7.14%. The SVM model did not make any wrong recognition for J-type discharges, which shows that the algorithm has a good credibility for J-type partial discharges. Compared with the SVM algorithm, the CNN network has the largest improvement in the recognition rate of K-type partial discharges. Compared with the CNN network, the DBN network has the largest improvement in the recognition of X-type partial discharges, and also has a certain improvement in the recognition of Q-type partial discharges. In the recognition results of the DBN network, there is no misidentification of J-type and K-type partial discharges, which shows that the network model has a high credibility for the recognition of these two types of partial discharges. Based on the accurate recognition rate of the DBN network, the PSO-DBN network further improved the recognition of X-type and K-type partial discharges by 3.7% and 3.85% respectively, and also had a high credibility for the recognition of J-type and K-type partial discharges.

[0137] By comparing the results of the three recognition models of BP, SVM, and CNN, it can be concluded that these three recognition models mainly misidentify X type and K type, followed by Q type, and almost no J type partial discharge misidentification. It can be seen from the PRPD spectrum of the four types of partial discharge that the J type has more obvious feature differences than the other three types, and the other three types have certain similarities in phase distribution, which increases the difficulty in recognition. However, the DBN network with a deep hierarchical structure can fully learn the correlation factors between more potential partial discharge features, so that it has a higher recognition ability for the three types of partial discharges. In addition, the particle swarm optimization algorithm is used in the present invention to optimize the weight parameters of the DBN network model, and a higher accurate recognition rate is obtained, which is more applicable than the first three networks.

[0138] The accuracy and overall accuracy of the DBN network model optimized by particle swarm optimization and the other four models for four typical partial discharge types are shown in the figure. Figure 7 As shown in the figure, it can be seen that in the identification of K-type partial discharge, PSO-DBN has the largest improvement compared with BP network, with an improvement value of more than 15%. The second largest improvement in accuracy is the identification of X-type partial discharge, with an improvement value of more than 10%. In the overall recognition rate, it has the largest improvement compared with BP network, followed by SVM, then CNN, and finally traditional DBN network. PSO-DBN is better than the other four algorithms in the four types of partial discharge and the overall recognition accuracy.

[0139] In the entire partial discharge identification and comparison process, the DBN network with stronger learning ability has better partial discharge identification effect than the traditional commonly used BP, SVM, CNN and other network algorithms, especially for partial discharge types with certain similarity characteristics. The PSO-DBN network model with network parameters optimized by particle swarm algorithm has a higher partial discharge accuracy recognition rate than the unoptimized DBN network.

[0140] Example 3

[0141] The above-mentioned embodiment 1 provides a GIS partial discharge pattern recognition method. Correspondingly, this embodiment provides a GIS partial discharge pattern recognition system. The system provided by this embodiment can implement the GIS partial discharge pattern recognition method of embodiment 1, and the system can be implemented by software, hardware, or a combination of software and hardware. For example, the system may include integrated or separate functional modules or functional units to perform the corresponding steps in each method of embodiment 1. Since the system of this embodiment is basically similar to the method embodiment, the description process of this embodiment is relatively simple. For relevant parts, please refer to the partial description of embodiment 1. The embodiment of the system provided by this embodiment is only illustrative.

[0142] The GIS partial discharge pattern recognition system provided in this embodiment includes:

[0143] The data acquisition module is used to select the on-site multi-platform GIS partial discharge monitoring data to form a training sample set;

[0144] The model training module is used to optimize the model parameters of the pre-established DBN network model using the improved PSO algorithm, and obtain the optimal DBN network model based on the training sample set data;

[0145] The partial discharge recognition module is used to perform partial discharge pattern recognition on the target GIS switch device based on the obtained optimal DBN network model to obtain the partial discharge pattern recognition result of the target GIS switch device.

[0146] Example 4

[0147] This embodiment provides a processing device corresponding to the GIS local discharge pattern recognition method provided in this embodiment 1. The processing device can be a processing device for a client, such as a mobile phone, a laptop computer, a tablet computer, a desktop computer, etc., to execute the method of embodiment 1.

[0148] The processing device includes a processor, a memory, a communication interface and a bus, and the processor, the memory and the communication interface are connected through the bus to complete mutual communication. The memory stores a computer program that can be run on the processor, and the processor executes the GIS partial discharge pattern recognition method provided in this embodiment 1 when running the computer program.

[0149] Preferably, the memory may be a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory.

[0150] Preferably, the processor may be a central processing unit (CPU), a digital signal processor (DSP) or other general-purpose processors of various types, which are not limited here.

[0151] Example 5

[0152] The GIS local discharge pattern recognition method of this embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the GIS local discharge pattern recognition method described in this embodiment 1 are loaded.

[0153] Computer readable storage media can be tangible devices that hold and store instructions used by instruction execution devices. Computer readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any combination thereof.

[0154] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0155] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0156] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0157] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0158] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A GIS partial discharge pattern recognition method, characterized in that: The following steps are involved: The on-site multi-platform GIS partial discharge monitoring data are selected to form a training sample set; The improved PSO algorithm is used to optimize the model parameters of the pre-established DBN network model, and the optimal DBN network model is obtained based on the training sample set data. Based on the obtained optimal DBN network model, partial discharge pattern recognition is performed on the target GIS switch equipment to obtain the partial discharge pattern recognition result of the target GIS switch equipment.

2. A GIS partial discharge pattern recognition method as claimed in claim 1, characterized in that: The on-site multi-platform GIS partial discharge monitoring data is selected to form a training sample set, including: Obtain partial discharge monitoring data of multi-platform GIS switchgear on site; Preprocess and normalize the acquired partial discharge monitoring data; The normalized partial discharge monitoring data is divided into a training set and a test set according to a preset ratio as a training sample set.

3. A GIS partial discharge pattern recognition method as claimed in claim 2, characterized in that: The partial discharge monitoring data include: skewness, correlation coefficient, discharge amount factor, asymmetry, steepness and number of local peaks.

4. A GIS partial discharge pattern recognition method as claimed in claim 2, characterized in that: The improved PSO algorithm is used to optimize the model parameters of the pre-established DBN network model, and the optimal DBN network model is obtained based on the training sample set data, including: The improved PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration times in the DBN network model to obtain the optimal particle vector. The optimal particle vector is set as the corresponding DBN network model parameter, the initial optimal DBN network model is established and trained using the training set data; The optimal DBN network model obtained through training is used to identify, classify and verify the test set data to obtain the optimal DBN network model that meets the preset accuracy.

5. A GIS partial discharge pattern recognition method as claimed in claim 4, characterized in that: The improved PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration number parameters in the DBN network model to obtain the optimal particle vector, including: Establish DBN network model; The optimized PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration times of the DBN network model to obtain the optimal particle vector.

6. A GIS partial discharge pattern recognition method as claimed in claim 5, characterized in that: The DBN network model includes: Input layer, first to third RBM layers, BP layer and output layer; The input layer is used to obtain input feature data; The first to third RBM layers are used to extract feature information of input partial discharge data and supervised reverse fine-tuning model parameters through forward unsupervised autonomous learning; The BP layer is used to reduce the error between the final recognition result and the actual result; The output layer is used to output annotated partial discharge data.

7. A GIS partial discharge pattern recognition method as claimed in claim 5, characterized in that: The optimized PSO algorithm is used to optimize the network structure, learning rate, and forward and reverse iteration number parameters of the DBN network model to obtain the optimal particle vector, including: ①Determine the fitness function of the particle swarm; ② Determine the relevant parameter range of the particle swarm and initialize the correlation coefficient of the particles; ③ Map the initialized DBN network model parameters to the dimensions of the particles; ④ Iterate and optimize the weight parameters of the DBN network model according to the current parameters of each particle, and calculate the fitness value of each particle after the kth iteration of value search; ⑤ Update the optimal value of each particle and population in the current iteration round according to the fitness value and extreme value before and after the particle optimization, and update the inertia coefficient and learning factor of the particle according to the current number of iterations k; ⑥ Use the updated particle coefficient to update the speed and position of all particles in the population, and determine whether the current iteration number in the particle optimization process is equal to the maximum iteration number of the particle set in advance: If the current iteration number of the particle k≠T, return to step ⑤ to continue the particle iterative optimization process; if the current iteration number k=T, the PSO algorithm iterative optimization ends and the optimal particle vector is obtained.

8. A GIS partial discharge pattern recognition system, characterized in that: include: The data acquisition module is used to select the on-site multi-platform GIS partial discharge monitoring data to form a training sample set; The model training module is used to optimize the model parameters of the pre-established DBN network model using the improved PSO algorithm, and obtain the optimal DBN network model based on the training sample set data; The partial discharge recognition module is used to perform partial discharge pattern recognition on the target GIS switch device based on the obtained optimal DBN network model to obtain the partial discharge pattern recognition result of the target GIS switch device.

9. A computer-readable storage medium storing one or more programs, characterized in that: The one or more programs include instructions, which, when executed by a computing device, cause the computing device to perform any one of the methods of claims 1 to 7.

10. A computing device, characterized in that: include: One or more processors and a memory, wherein the memory stores one or more programs and is configured to be executed by the one or more processors, wherein the one or more programs include instructions for executing any one of the methods described in claims 1 to 7.

Images