Online monitoring data analysis method and system for dissolved gas in oil of oil-immersed transformer

Abstract

The invention discloses a GA-CNN-based online monitoring data analysis method, system, equipment, medium and program for dissolved gas in oil of an oil-immersed transformer, and belongs to the technical field of transformer detection. The method comprises the following steps: preprocessing historical fault data and normal data of a transformer to be analyzed to obtain a sample data set; and inputting the sample data set into a pre-trained CNN transformer fault diagnosis model, and performing analysis to obtain an analysis result. According to the method, different types of transformers can be identified, and the identification accuracy of overheating and discharge defects of the oil-immersed transformer is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer detection, and in particular to a method, system, equipment, medium and program for online monitoring data analysis of dissolved gas in oil of an oil-immersed transformer based on GA-CNN. Background Art

[0002] For hydropower plants, transformers are not only the key hub for power conversion and transmission, but also the core equipment that is indispensable to ensure the stable operation of the power system. Its health directly affects the production efficiency, economic benefits and power supply reliability of the hydropower plant. Therefore, accurate fault identification and timely maintenance of transformers are of great significance to ensure the normal operation of hydropower plants.

[0003] Oil-immersed transformers are an important part of the transformer family. The liquid insulating material inside them, namely insulating oil, plays a vital role in the long-term operation of the transformer. Insulating oil can not only effectively isolate electrical components and prevent short-circuit accidents, but also help the transformer maintain a suitable operating temperature through its good heat dissipation performance. However, it is this seemingly calm internal environment that actually hides many potential risks.

[0004] In complex working environments, such as high temperature, strong electric field and possible catalysts, the insulating oil of oil-immersed transformers will gradually undergo chemical changes and undergo electrical aging and thermal aging. These aging processes will cause the molecular chains of the insulating oil to break and produce a series of specific dissolved gases, such as hydrogen, methane, ethane, ethylene, acetylene, etc. The types, contents and changing trends of these gases are like a mirror, reflecting the health status and fault types inside the transformer.

[0005] Specifically, different types of faults are often accompanied by the generation of specific gas combinations. For example, partial discharge may lead to a significant increase in hydrogen content; overheating faults may cause an increase in alkane gases such as methane and ethane; and arc discharges will produce a large amount of acetylene gas. Through qualitative and quantitative analysis of dissolved gases in insulating oil, technicians can preliminarily determine the fault type, fault location and degree of development of the transformer, providing a strong basis for subsequent maintenance decisions.

[0006] At present, the main traditional methods for oil-immersed transformer fault diagnosis are the three-check method, characteristic gas method, David triangle method and three ratio method. The above methods are based on experience and rules or criteria formulated by experts to determine the type of internal fault of the transformer. Due to the differences in internal structure, insulating oil composition, volume and voltage level of transformers from different manufacturers and models, coupled with the complex internal fault mechanism of the transformer, a single and strict judgment rule cannot effectively distinguish the fault, and it is easy to miss or misjudge. Summary of the invention

[0007] In view of the differences in the internal structure, insulating oil composition, volume and voltage level of transformers of different manufacturers and models in the prior art, and the complex internal fault mechanism of the transformer, the existing judgment rules cannot effectively distinguish transformer faults, and are prone to missed judgments and misjudgments. The present invention provides an online monitoring data analysis method for dissolved gas in oil of an oil-immersed transformer, which can identify different types of transformers and effectively improve the recognition accuracy of overheating and discharge defects of oil-immersed transformers.

[0008] In order to achieve the above object, the present invention provides the following technical solutions.

[0009] In a first aspect, the present invention provides a method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil, comprising: The historical fault data and normal data of the transformer to be analyzed are preprocessed to obtain a sample data set; Input the sample data set into the pre-trained CNN transformer fault diagnosis model for analysis to obtain the analysis results; The training method of the pre-trained CNN transformer fault diagnosis model includes: Collect historical fault data and normal data of transformers, and perform preprocessing to obtain sample data sets; The training data set in the sample data set is input into the transformer fault diagnosis model of CNN, and the hyperparameters are optimized using genetic algorithm to obtain the optimal CNN hyperparameters; The CNN transformer fault diagnosis model is retrained using the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model.

[0010] As a further improvement of the present invention, the historical fault data and normal data of the transformer are collected and preprocessed to obtain a sample data set, including: According to the data collected from the transformer online monitoring system and offline oiling experiments, the real-time concentration data of characteristic gases such as hydrogen, methane, ethylene, and acetylene in transformer oil are obtained; Analyze and judge the real-time concentration data of characteristic gases such as hydrogen, methane, ethylene, and acetylene in transformer oil to obtain historical fault data and normal data of the transformer; The isolation forest algorithm is used to preprocess the historical fault data and normal data of the transformer, and the outliers are eliminated. The historical fault data and normal data of the transformer after elimination are subjected to data standardization to obtain a sample data set.

[0011] As a further improvement of the present invention, the isolation forest algorithm is used to pre-process the historical fault data and normal data of the transformer to remove abnormal values, including: A feature is randomly selected from the transformer historical fault data and normal data, and a value between the maximum value and the minimum value of the feature is randomly selected as the partition threshold; According to the selected features and thresholds, the data set is divided into two subspaces, left and right; each subspace is divided recursively until the tree reaches the preset maximum height or there is only one data point in the subspace; until several independent isolated trees are constructed to form an isolation forest; Calculate the anomaly score of a data point based on its path length in multiple isolated trees; A threshold is set based on the anomaly score, and data points with scores above the threshold are considered outliers.

[0012] As a further improvement of the present invention, the training data set in the sample data set is input into the transformer fault diagnosis model of CNN, and the hyperparameters are optimized using a genetic algorithm to obtain the optimal CNN hyperparameters, including: Input the training data set in the sample data set into the transformer fault diagnosis model of CNN and select the hyperparameters; Convert the selected hyperparameters into genetic encoding form and generate a set of random initial solutions as the first generation population; By defining a fitness function to evaluate the performance of each individual in the first generation population, individuals are selected for reproduction according to the fitness value to obtain a new generation population; The individual with the highest fitness in the new generation population is the optimal CNN hyperparameter.

[0013] As a further improvement of the present invention, the step of evaluating the performance of each individual in the first generation population by defining a fitness function, selecting individuals for reproduction according to the fitness value, and obtaining a new generation population comprises: The fitness function is defined to evaluate the performance of each individual in the first generation population and obtain the fitness value.

[0014] Select individuals according to their fitness values, perform crossover operations on the selected individuals, and randomly change the genes of the selected individuals to introduce new genetic information; Repeat the selection, crossover and mutation operations to generate several new generations of populations until the overall fitness of the population reaches the set stop condition; In each generation, the population is evaluated according to the fitness function and individuals with higher fitness are retained.

[0015] As a further improvement of the present invention, the transformer fault diagnosis model of CNN is retrained using the optimal CNN hyperparameters to obtain the trained transformer fault diagnosis model of CNN, including: Retraining the CNN transformer fault diagnosis model according to the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model; The performance of the trained CNN transformer fault diagnosis model is evaluated based on the test set of the sample data set; If the performance evaluation result meets the set conditions, the training of the CNN transformer fault diagnosis model is completed, and the trained CNN transformer fault diagnosis model is obtained; If the performance evaluation result does not meet the set conditions, the CNN transformer fault diagnosis model is retrained until the performance evaluation result meets the set conditions.

[0016] In a second aspect, the present invention provides an online monitoring data analysis system for dissolved gas in oil of an oil-immersed transformer, comprising: Preprocessing module: used to preprocess the historical fault data and normal data of the transformer to be analyzed to obtain a sample data set; Identification module: used to input the sample data set into the pre-trained CNN transformer fault diagnosis model for analysis and obtain the analysis results; The training method of the pre-trained CNN transformer fault diagnosis model includes: Data preprocessing module: used to collect historical fault data and normal data of transformers, and perform preprocessing to obtain sample data sets; Obtain CNN hyperparameter module: used to input the training data set in the sample data set into the transformer fault diagnosis model of CNN, and use genetic algorithm to optimize the hyperparameters to obtain the optimal CNN hyperparameters; Training model module: used to retrain the CNN transformer fault diagnosis model using the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model.

[0017] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of a method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil when executing the computer program.

[0018] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil are implemented.

[0019] In a fifth aspect, the present invention provides a computer program product, characterized in that it includes computer instructions, which, when executed by a processor, implement the steps of the method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil.

[0020] Compared with the prior art, the present invention has the following beneficial effects: The present invention introduces a pre-trained convolutional neural network (CNN) model to conduct deep learning and analysis on the historical fault data and normal data of the transformer, and can more accurately identify the overheating and discharge defects inside the transformer. Compared with the traditional judgment rules, the CNN model can automatically extract deep features in the data, thereby improving the accuracy of recognition. Considering that transformers of different manufacturers and models have differences in internal structure, insulating oil composition, volume size and voltage level, the method of the present invention trains the CNN model to enable it to learn these differences and effectively distinguish them. Therefore, the method can be applied to different types of transformers and has strong adaptability. Secondly, due to the complex internal fault mechanism of the transformer, the traditional judgment rules are prone to missed judgment and misjudgment. The method of the present invention can more comprehensively analyze the data of dissolved gas in transformer oil through a deep learning algorithm, thereby reducing the possibility of missed judgment and misjudgment. The method of the present invention supports online monitoring of the data of dissolved gas in transformer oil and performs real-time analysis. This helps to timely discover potential faults of the transformer, provide timely warnings for maintenance and replacement, and thus ensure the stable operation of the power system. At the same time, in the process of training the CNN model, the present invention uses a genetic algorithm to optimize the hyperparameters to obtain the optimal CNN hyperparameters. This helps to improve the performance of the model so that it can better adapt to the fault diagnosis task of the transformer. Therefore, the present invention can greatly shorten the diagnosis time and improve the diagnosis efficiency by using the trained CNN model for fault diagnosis. This is of great significance for the maintenance and operation management of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way. In the drawings: Figure 1 A schematic flow chart of an online monitoring data analysis method for dissolved gas in oil-immersed transformer oil according to the present invention; Figure 2 A flowchart of a training method for a transformer fault diagnosis model of a pre-trained CNN of the present invention; Figure 3 The present invention is a flow chart of a method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil based on GA-CNN; Figure 4 This is a scatter diagram of the original data; Figure 5 This is a schematic diagram of data scatter after outliers are removed; Figure 6 Schematic diagram of the isolation forest algorithm; Figure 7 Schematic diagram of convolution operation; Figure 8 are Tanh and Sigmoid function images; Fig. 9 is the ReLU function image; Fig.10 Schematic diagram of pooling; Fig.11 It is the fully connected layer and the full output layer; Fig.12 It is the flow chart of back propagation; Fig.13 A bar chart comparing the accuracy of various algorithm results; Fig.14 This is a comparison chart of the GA-CNN algorithm prediction and actual results; Fig.15 It is a structural schematic diagram of an online monitoring data analysis system for dissolved gas in oil of an oil-immersed transformer according to the present invention; Fig.16 A schematic diagram of the structure of a training system for a transformer fault diagnosis model of a pre-trained CNN of the present invention; Fig.17 Schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the present invention will be clearly and completely described below in conjunction with the drawings in the present invention. The described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0024] In view of the differences in internal structure, insulating oil composition, volume, voltage level, etc. of transformers of different manufacturers and models in the prior art, and the complex internal fault mechanism of the transformer, the existing judgment rules cannot effectively distinguish transformer faults, and are prone to missed judgments and misjudgments. The present invention provides an online monitoring data analysis method for dissolved gas in oil of oil-immersed transformers, such as Figure 1 As shown, the method includes: S100: preprocessing the historical fault data and normal data of the transformer to be analyzed to obtain a sample data set; S200: input the sample data set into the pre-trained CNN transformer fault diagnosis model, analyze it, and obtain the analysis result; The training method of the pre-trained CNN transformer fault diagnosis model includes: S210: Collect historical fault data and normal data of the transformer, and perform preprocessing to obtain a sample data set; S220: inputting the training data set in the sample data set into the transformer fault diagnosis model of CNN, and optimizing the hyperparameters using a genetic algorithm to obtain the optimal CNN hyperparameters; S230: Retrain the transformer fault diagnosis model of CNN using the optimal CNN hyperparameters to obtain the trained transformer fault diagnosis model of CNN.

[0025] The present invention can identify different types of transformers, and effectively improves the recognition accuracy of overheating and discharge defects of oil-immersed transformers.

[0026] The present invention is further explained below in conjunction with the specific drawings.

[0027] like Figure 3 As shown, the present invention provides an online monitoring data analysis method for dissolved gas in oil of an oil-immersed transformer based on GA-CNN, which effectively improves the recognition accuracy of overheating and discharge defects of the oil-immersed transformer, and specifically includes the following steps.

[0028] S1: Obtain raw data and collect transformer online monitoring system and offline oiling test data.

[0029] Specifically, five characteristic gases, hydrogen (H2), methane (CH4), ethylene (C2H4), and acetylene (C2H2), were selected as sample features according to the "Guidelines for Analysis and Judgment of Dissolved Gases in Transformer Oil". The historical fault data of the transformer and some normal data were collected offline to form a data set.

[0030] S2: Data preprocessing, abnormal sample cleaning, and cleaning of abnormal samples based on isolation forest; standardization of data to ensure consistency in the numerical range of different features, which is conducive to model training; processing of transformer fault labels, encoding of five equipment states: normal, low energy discharge, high energy discharge, medium and low temperature overheating, and high temperature overheating, and establishing a coding and state mapping table.

[0031] Specifically, the isolation forest algorithm is used to detect outliers.

[0032] The algorithm steps are as follows: (1) Randomly select features and thresholds: ① Randomly select a feature from the given dataset.

[0033] ② Randomly select a value of the feature between the maximum and minimum values ​​as the partition threshold.

[0034] (2) Constructing an isolation tree: ① Divide the data set according to the selected features and thresholds, and assign the data points to the left or right subspace.

[0035] ② Recursively divide the subspace until one of the following conditions is met: (a) The tree reaches a preset maximum height (the depth limit of the tree).

[0036] (b) There is only one data point in the subspace, namely the leaf node.

[0037] (3) Construct multiple isolated trees: Repeat steps 1 and 2 to construct multiple independent isolated trees.

[0038] (4) Outlier detection: ① For each data point, the degree of outlier is evaluated by calculating its path length in each isolated tree.

[0039] ②Outliers usually have shorter average path lengths because they are easier to isolate.

[0040] (5) Calculate the anomaly score: ① Calculate the anomaly score based on the path length of the data point in multiple isolated trees.

[0041] ②The shorter the path length, the higher the anomaly score, which means the data point is more likely to be an outlier.

[0042] (6) Determine outliers: ① Set the threshold according to the anomaly score and judge the outliers of the data points.

[0043] ②Data points with scores higher than the threshold are considered outliers.

[0044] The outlier detection results are as follows Figure 4 and Figure 5 As shown in the figure, a total of 18 outliers were removed.

[0045] Furthermore, the data is standardized and the calculation formula is as follows:

[0046] in is the mean of the eigenvectors, is the variance of the eigenvector.

[0047] Furthermore, for the processing of transformer fault labels, the low temperature overheating fault and the medium temperature overheating fault are first merged into the medium and low temperature overheating fault. Finally, the five fault types of normal, low temperature overheating, high temperature overheating, low energy discharge, and high energy discharge are digitally encoded as sample labels. The encoding rules are shown in Table 1.

[0048] Table 1 Classification of transformer electrical faults

[0049] S3: Design and build a CNN-based transformer fault diagnosis model, randomly divide all samples into training data sets and test data sets in a ratio of 8:2, and use the training data to train the model.

[0050] Specifically, the input layer, convolution layer, activation function, pooling layer, fully connected layer and output layer of the CNN network are constructed. The specific principles of the CNN algorithm are as follows: Convolutional Neural Network (CNN) is a deep learning model specifically designed to process data with a grid-like structure. CNN is mainly composed of convolutional layers, pooling layers, and fully connected layers. Its network structure is as follows: Figure 6 shown.

[0051] The convolution layer extracts features by applying convolution kernels (filters) to the input data. Each convolution kernel can detect a specific feature in the input data, such as edges, textures, etc. The convolution operation can preserve the spatial structure of the features, reduce the number of parameters, and share weights to improve the generalization ability of the model.

[0052] In the computer field, each image can be regarded as a matrix composed of several pixel values ​​between 0 and 255. The image is generally normalized before being input into the neural network model. The convolution layer performs convolution operations on the data through the convolution kernel to obtain its features. For a deep CNN structure with multiple convolution layers, the subsequent convolution layer convolves the feature map output by the previous convolution layer again to extract its deeper features. The convolution layer consists of multiple convolution kernels of the same size. The more common convolution kernel sizes are 5×5 and 3×3. The area covered by the input image increases with the size of the convolution kernel during each convolution operation.

[0053] In the computer field, each image can be regarded as a matrix composed of several pixel values ​​between 0 and 255. The image is generally normalized before being input into the neural network model. The convolution layer performs convolution operations on the data through the convolution kernel to obtain its features. For a deep CNN structure with multiple convolution layers, the subsequent convolution layer convolves the feature map output by the previous convolution layer again to extract its deeper features. The convolution layer consists of multiple convolution kernels of the same size. The more common convolution kernel sizes are 5×5 and 3×3. The area covered by the input image increases with the size of the convolution kernel during each convolution operation.

[0054] like Figure 7 As shown in the figure, taking a one-channel grayscale image of size 5×5×1 as an example, its pixel values ​​are only 0 and 1, and the convolution kernel size is 3×3×1. The convolution operation can be understood as follows: the convolution kernel can be regarded as a window with a moving step of 1. First, the inner product of each value and the pixel point of the input data is obtained, and then the sum is calculated to finally obtain a feature map of size 3×3×1. In practical applications, multiple convolution kernels are often selected. The number of channels of the convolution kernel is equal to the number of channels of the input image, and the number of channels of the output feature map is equal to the number of convolution kernels.

[0055] The moving step refers to the distance the convolution kernel slides each time. If the moving step is 1, it means that the convolution kernel moves one pixel unit in one direction. The longer the moving step, the smaller the overlap of the areas involved in the convolution operation, which can reduce the amount of redundant network calculations. However, if the moving step is too large, it may also lead to insufficient feature extraction. The moving step is generally set to 1 or 2. Since the frequency of the edge of the input image participating in the convolution operation is much lower than that of the central area, in order to avoid losing the edge features of the input image, it is necessary to fill a layer of blank pixels at the edge of the input image. There are two types of filling methods: (1) Same, fill the blank pixels around the input image, and if the moving step = 1, this type of filling method can ensure that the input image and the output feature map are the same size. (2) Valid, which means no filling. The size relationship between the input image and the output image after the convolution operation is as shown in formula (1) (1) in, Represents the length and width of the input image, Represents the length and width of the input feature map, Represents Padding, Represents the length and width of the convolution kernel, Represents the moving step length.

[0056] Table 2 Common activation functions

[0057] After the convolution operation, the pixel values ​​of the output feature map need to be nonlinearly mapped to map the features of the high-dimensional space to the low-dimensional space to increase its linear separability. In addition, since CNN often targets nonlinear problems, a nonlinear activation function is selected, as shown in Table 2. The images of Tanh and Sigmoid functions are shown in Figure 8 shown.

[0058] The ReLU function has the advantages of sparse activation and simple calculation, and can effectively prevent gradient disappearance. Therefore, it is most frequently used in CNN. The image of the ReLU function is as follows: Fig. 9 shown.

[0059] In summary, the forward propagation calculation of the convolutional layer can be summarized as follows: (2) Among them, * represents the convolution operator symbol, represents a set of input feature maps, Indicates network Tier The convolution kernel weights, represents the bias, Indicates network Layer The input feature map of the convolution kernel, is the activation function, Representative Layer An input feature map.

[0060] (2) Pooling layer The pooling layer is used to reduce the spatial dimension of the feature map, reduce the amount of calculation, and increase translation invariance. Common pooling operations include maximum pooling and average pooling. Fig.10 The figure is a schematic diagram of pooling. Average pooling and maximum pooling are the two main pooling methods. The output of the former is the average value of the pixels in the square box, and the output of the latter is the maximum value of the pixels in the square box. The calculation method of pooling can be expressed as:

[0061] in, and represent the addition and multiplication parameters respectively, is the downsampling calculation formula. According to the texture feature analysis research in the direction of machine vision, maximum pooling has a good ability to extract periodic signal features. Therefore, the models proposed in this paper all use maximum pooling.

[0062] Similar to the convolution layer, the pooling layer has several hyperparameters that need to be set, such as the moving step, padding, and sampling window size. In practical applications, most CNN models set the moving step to 2 and the sampling window size to 2×2. Since the length and width of the feature map output by the convolution layer are generally even numbers, the padding mode is set to Valid. The size of the feature map after this type of pooling operation will be halved, which greatly reduces the parameters and calculation amount of the model. It should be noted that there are no parameters in the pooling layer that need to be iteratively updated during the training process.

[0063] (3) Fully connected layer CNN achieves feature extraction through the above-mentioned convolutional layer and pooling layer, and the fully connected layer can be regarded as a feature classification layer, just like a typical feedforward neural network. After the convolutional layer and pooling layer extract the features of the image, the features are flattened and finally input into the fully connected layer for model learning. Fig.11 The fully connected layer and the output layer are schematic diagrams. The fully connected layer, like the hidden layer of the neural network, plays a buffering role between the feature map and the output layer. Its expression is as follows: (3) (4) Output layer The output layer of CNN varies for different tasks. When dealing with image classification problems, a Softmax classifier is usually set at the output layer of CNN, which can convert the k-dimensional input vector into a vector in another feature space. , and the vector The sum of each absolute value in is equal to 1. The output of the Softmax classifier is shown as follows: (4) It can be seen that its output represents the possibility of the category corresponding to each input sample. The larger the output value, the greater the probability that the predicted category is the true category.

[0064] (5) Loss function The loss function can measure the difference between the output of the deep learning model and the probability distribution of the true category of the sample. The smaller the loss function value, the better the training effect of the network and the higher the prediction accuracy. As shown in formula (5), CNN generally uses the cross entropy loss function.

[0065] (5) in, and They represent the probability distribution of the actual sample category and the model output result respectively.

[0066] (6) Back propagation There are two main steps for CNN to update parameters: forward propagation and back propagation. Forward propagation refers to the process from input image to feature extraction, and then to classification and output classification results. Its function is to pass the features of the image layer by layer to complete the classification of the image. During forward propagation, the classification result will be compared with the actual label of the image, and the difference will be reflected in the loss function. After that, the loss function will pass the classification error to the previous layers of the network layer by layer through the back propagation method, adjust the weight and bias of each neural node, and minimize the classification error of the model. The process of back propagation is as follows: Fig.12 shown.

[0067] The gradient descent method is generally used to calculate the minimum value of the loss function in the deep learning model. Its iterative formula is: (6) in, is the negative gradient direction, is the moving step in the gradient direction. The gradient direction can be obtained by calculating the partial derivative of the function. is the current function value. is the target function value.

[0068] The convolutional neural network parameter update process can be summarized as follows: Assume that the objective function of the network is , and are the weights and biases of the network respectively. First, the output of each layer of the network is obtained according to the network forward propagation algorithm , and then the output of the last layer is expressed as , the objective function About Network Tier Neuron output The partial derivative of is defined as , and The calculation formulas are as follows: (7) (8) Objective Function about and The partial derivative of about and The gradient of can be found using the chain rule: (9) (10) Then calculate the loss function About Network ) layer output value and hidden layer activation value Partial derivative of: (11) (12) Calculate After the value is obtained, the partial derivative of the objective function with respect to the weights and biases of the previous layer is calculated again. Finally, the parameters are updated using the gradient descent method as follows: (13) (14) is the updated weight of CNN, is the updated bias of CNN, Represents the learning rate, that is, the moving step size.

[0069] S4: GA-based CNN hyperparameter optimization, using genetic algorithm to search in the hyperparameter space to find the best hyperparameter combination.

[0070] The specific algorithm implementation steps are as follows: (1) Convert the selected hyperparameters into genetic encoding form. Each hyperparameter can be mapped to one or more genes, and the entire model configuration can be represented as a chromosome. Generate a set of random initial solutions (i.e., a set of CNN hyperparameter configurations) as the first generation population. The population size should be large enough to ensure the diversity of the genetic algorithm.

[0071] (2) Define a fitness function to evaluate the performance of each individual. In transformer fault diagnosis, the fitness function can be the model's accuracy, F1 score, or other performance indicators. Usually, this function is calculated based on the performance on the validation set.

[0072] (3) Select individuals for reproduction based on fitness values. Common selection methods include roulette selection and tournament selection. Individuals with high fitness have a higher probability of being selected. The selected individuals are subjected to a crossover operation, that is, some genes are exchanged to produce new individuals. The crossover operation can simulate the gene recombination process in biological evolution. The genes of individuals are randomly changed with a certain probability to introduce new genetic information. The mutation operation can increase the diversity of the population and help avoid local optimal solutions.

[0073] (4) Repeat the selection, crossover and mutation operations to generate a new generation of population. Each generation evaluates the population according to the fitness function and retains individuals with higher fitness. As the number of iterations increases, the overall fitness of the population will gradually increase until the preset stopping condition (such as the maximum number of iterations, fitness threshold, etc.) is reached.

[0074] (5) When the stopping condition is reached, the genetic algorithm terminates. At this point, the individual with the highest fitness in the population is the optimal CNN hyperparameter configuration. Use this configuration to retrain the CNN model and evaluate its performance on the test set.

[0075] S5: Use the test data to verify the optimized CNN model and evaluate the model.

[0076] Specifically, in the theoretical verification phase, this project trained and verified the standard DGA dataset, and used CNN, RF, SVM, and GA-CNN algorithms to perform model training, evaluation, and comparative verification. The evaluation results are as follows: Fig.13 shown.

[0077] Comparing the GA-CNN proposed in this project with the CNN, RF, and SVM algorithms, it can be seen that the accuracy of GA-CNN is better than the other three algorithms in terms of both single fault recognition accuracy and comprehensive accuracy. The comprehensive accuracy of GA-CNN is 91.18%, CNN is 85.29%, and RF and SVM are 73.53%. Comparing the prediction results of the GA-CNN algorithm with the actual results, as shown in the figure. Fig.14 shown.

[0078] As can be seen from the above figure, among the six data that GA-CNN misjudged, two were normal data judged as faulty data, and four were fault type misjudgments. There was no case of missed faults, which was in line with the expectations for fault judgment in this project.

[0079] Before the implementation of this project, the oil chromatography monitoring system used by the power plant mainly focused on the following functions: online detection, data collection, and fault diagnosis. Fault diagnosis is mainly carried out through the modified three-ratio method, David triangle method, and cube diagram method. This project uses the GA-CNN algorithm to analyze dissolved gas in oil. Compared with the above three methods, the GA-CNN algorithm has obvious advantages in the accuracy of fault identification.

[0080] Specifically, the GA-CNN algorithm combines the global search capability of the genetic algorithm (GA) and the deep learning capability of the convolutional neural network (CNN), and can more accurately extract and analyze the characteristics of dissolved gas in oil. This project uses the DGA data set to continue testing the improved three-ratio method, David triangle method, cube diagram method, and GA-CNN method for fault recognition accuracy, and the results are shown in Table 3 below.

[0081] Table 3 Fault identification accuracy

[0082] As shown in Table 3, in the test, the GA-CNN algorithm achieved a fault recognition accuracy of 91.2%, while the recognition accuracy of the traditional modified three-ratio method, David triangle method and cube diagram method used by the power plant in the oil chromatography online monitoring system was 83.2%, 84.3% and 82.4% respectively. This shows that the GA-CNN algorithm used in this project has better fault recognition ability than the other three traditional methods under complex working conditions, and improves the fault judgment accuracy by 6.9%, proving the superiority of the GA-CNN algorithm in the online monitoring and analysis of dissolved gas in oil.

[0083] In summary, the method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil based on GA-CNN in the present invention, combined with the method of genetic algorithm (GA) and convolutional neural network (CNN), can effectively improve the accuracy and efficiency of transformer fault diagnosis, and provide new technical means for intelligent maintenance and fault prediction of power plants, which is of great significance for safe operation of equipment, life extension, economic operation of power plants, and condition maintenance. The method obtains characteristic information of potential faults inside the transformer by monitoring the gas content data in the insulating oil. The hyperparameters of the CNN model are optimized by genetic algorithm to improve the model performance and generalization ability, thereby realizing accurate diagnosis of transformer faults. The genetic algorithm simulates the biological evolution process, and searches for the optimal hyperparameter combination through selection, crossover and mutation operations, so as to better adapt to the characteristics of the DGA data set. At the same time, the convolutional neural network can effectively extract features from complex data, and perform fault classification prediction, providing reliable support for transformer maintenance.

[0084] The second object of the present invention is to provide an online monitoring data analysis system for dissolved gas in oil of an oil-immersed transformer. Fig.15 As shown, including: Preprocessing module 100: used to preprocess the historical fault data and normal data of the transformer to be analyzed to obtain a sample data set; Identification module 200: used to input the sample data set into the pre-trained CNN transformer fault diagnosis model, analyze it, and obtain the analysis result; like Fig.16 As shown, the training method of the pre-trained CNN transformer fault diagnosis model includes: Data preprocessing module 210: used to collect historical fault data and normal data of the transformer, and perform preprocessing to obtain a sample data set; Obtaining CNN hyperparameter module 220: used to input the training data set in the sample data set into the transformer fault diagnosis model of CNN, and use genetic algorithm to optimize the hyperparameters to obtain the optimal CNN hyperparameters; The training model module 230 is used to retrain the CNN transformer fault diagnosis model using the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model.

[0085] like Fig.17 As shown, the third object of the present invention is to provide an electronic device, the electronic device comprising: a processor 501, a memory 502 and a display screen 503. Among them, the memory 502 and the display screen 503 are connected to the processor 501, such as through a bus 504. Optionally, the electronic device may also include a transceiver 505. It should be noted that in actual applications, the transceiver 505 is not limited to one, and the structure of the electronic device does not constitute a limitation on the embodiments of the present application.

[0086] Processor 501 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array) or other programmable logic devices, transistor logic devices, hardware components or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of this application. Processor 501 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0087] The bus 504 may include a path to transmit information between the above components. The bus 504 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus 504 may be divided into an address bus, a data bus, a control bus, etc.

[0088] The memory 502 may be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory) or other optical disk storage, optical disk storage (including compressed optical disk, laser disk, optical disk, digital versatile disk, Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0089] The memory 502 is used to store application code for executing the solution of the present application, and the execution is controlled by the processor 501. The processor 501 is used to execute the application code stored in the memory 502 to implement the contents shown in the above method embodiment.

[0090] Fig.17 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0091] The fourth object of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is stored thereon, and when the program is executed by a processor, the aforementioned Figures 1 to 3 The various processes of the method embodiment shown in the figure include, for example, a memory including instructions, and the instructions can be executed by a processor of an electronic device to complete the method.

[0092] A computer-readable storage medium may be a tangible device that holds and stores instructions used by an instruction execution device. A computer-readable storage medium may be, but is not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof. Specifically, a computer-readable storage medium may be a portable computer disk, a hard disk, a USB flash drive, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a lectern random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, an optical disk, a magnetic disk, a mechanical encoding device, and any combination thereof.

[0093] A fifth object of the present invention is to provide a computer program product, comprising computer instructions, which, when executed by a processor, implement the above Figures 1 to 3 The various processes of the method embodiment shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0094] Many embodiments and many applications beyond the examples provided will be apparent to those skilled in the art upon reading the above description. Therefore, the scope of the present teachings should not be determined with reference to the above description, but rather with reference to the preceding claims and the full scope of equivalents to which such claims are entitled. For the purpose of comprehensiveness, all articles and references, including disclosures of patent applications and publications, are incorporated herein by reference. The omission of any aspect of the subject matter disclosed herein in the preceding claims is not intended to be a disclaimer of such subject matter, nor should it be considered that the applicants did not consider such subject matter to be part of the disclosed inventive subject matter.

[0095] The above content is a further detailed description of the present invention. It cannot be determined that the specific implementation methods of the present invention are limited to these. For ordinary technicians in the technical field to which the present invention belongs, without departing from the concept of the present invention, they can also make several simple deductions or substitutions, which should be regarded as belonging to the protection scope of the present invention determined by the submitted claims.

Claims

1. A method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil, characterized in that: include: The historical fault data and normal data of the transformer to be analyzed are preprocessed to obtain a sample data set; Input the sample data set into the pre-trained CNN transformer fault diagnosis model for analysis to obtain the analysis results; The training method of the pre-trained CNN transformer fault diagnosis model includes: Collect historical fault data and normal data of transformers, and perform preprocessing to obtain sample data sets; The training data set in the sample data set is input into the transformer fault diagnosis model of CNN, and the hyperparameters are optimized using genetic algorithm to obtain the optimal CNN hyperparameters; The CNN transformer fault diagnosis model is retrained using the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model.

2. The method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil according to claim 1, characterized in that: The transformer historical fault data and normal data are collected and preprocessed to obtain a sample data set, including: According to the data collected from the transformer online monitoring system and offline oiling experiments, the real-time concentration data of characteristic gases such as hydrogen, methane, ethylene, and acetylene in transformer oil are obtained; Analyze and judge the real-time concentration data of characteristic gases such as hydrogen, methane, ethylene, and acetylene in transformer oil to obtain historical fault data and normal data of the transformer; The isolation forest algorithm is used to preprocess the historical fault data and normal data of the transformer, and the outliers are eliminated. The historical fault data and normal data of the transformer after elimination are subjected to data standardization to obtain a sample data set.

3. The method for analyzing online data of dissolved gas in oil-immersed transformer oil according to claim 2, characterized in that: The isolation forest algorithm is used to pre-process the transformer historical fault data and normal data to remove abnormal values, including: A feature is randomly selected from the transformer historical fault data and normal data, and a value between the maximum value and the minimum value of the feature is randomly selected as the partition threshold; According to the selected features and thresholds, the data set is divided into two subspaces, left and right; each subspace is divided recursively until the tree reaches the preset maximum height or there is only one data point in the subspace; until several independent isolated trees are constructed to form an isolation forest; Calculate the anomaly score of a data point based on its path length in multiple isolated trees; A threshold is set based on the anomaly score, and data points with scores above the threshold are considered outliers.

4. The method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil according to claim 1, characterized in that: The training data set in the sample data set is input into the transformer fault diagnosis model of CNN, and the hyperparameters are optimized using a genetic algorithm to obtain the optimal CNN hyperparameters, including: Input the training data set in the sample data set into the transformer fault diagnosis model of CNN and select the hyperparameters; Convert the selected hyperparameters into genetic encoding form and generate a set of random initial solutions as the first generation population; By defining a fitness function to evaluate the performance of each individual in the first generation population, individuals are selected for reproduction according to the fitness value to obtain a new generation population; The individual with the highest fitness in the new generation population is the optimal CNN hyperparameter.

5. The method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil according to claim 4, characterized in that: The method of evaluating the performance of each individual in the first generation population by defining a fitness function and selecting individuals for reproduction according to the fitness value to obtain a new generation population includes: By defining the fitness function to evaluate the performance of each individual in the first generation population, the fitness value is obtained; Select individuals according to their fitness values, perform crossover operations on the selected individuals, and randomly change the genes of the selected individuals to introduce new genetic information; Repeat the selection, crossover and mutation operations to generate several new generations of populations until the overall fitness of the population reaches the set stop condition; In each generation, the population is evaluated according to the fitness function and individuals with higher fitness are retained.

6. The method for analyzing online monitoring data of dissolved gas in oil-immersed transformer oil according to claim 1, characterized in that: The transformer fault diagnosis model of CNN is retrained using the optimal CNN hyperparameters to obtain the trained transformer fault diagnosis model of CNN, including: Retraining the CNN transformer fault diagnosis model according to the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model; The performance of the trained CNN transformer fault diagnosis model is evaluated based on the test set of the sample data set; If the performance evaluation result meets the set conditions, the training of the CNN transformer fault diagnosis model is completed, and the trained CNN transformer fault diagnosis model is obtained; If the performance evaluation result does not meet the set conditions, the CNN transformer fault diagnosis model is retrained until the performance evaluation result meets the set conditions.

7. An online monitoring data analysis system for dissolved gas in oil-immersed transformer oil, characterized in that: include: Preprocessing module: used to preprocess the historical fault data and normal data of the transformer to be analyzed to obtain a sample data set; Identification module: used to input the sample data set into the pre-trained CNN transformer fault diagnosis model for analysis and obtain the analysis results; The training method of the pre-trained CNN transformer fault diagnosis model includes: Data preprocessing module: used to collect historical fault data and normal data of transformers, and perform preprocessing to obtain sample data sets; Obtain CNN hyperparameter module: used to input the training data set in the sample data set into the transformer fault diagnosis model of CNN, and use genetic algorithm to optimize the hyperparameters to obtain the optimal CNN hyperparameters; Training model module: used to retrain the CNN transformer fault diagnosis model using the optimal CNN hyperparameters to obtain the trained CNN transformer fault diagnosis model.

8. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil as described in any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil according to any one of claims 1 to 6 are implemented.

10. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the steps of the method for online monitoring data analysis of dissolved gas in oil-immersed transformer oil as described in any one of claims 1 to 6.

Images