Transformer fault diagnosis method and system based on dissolved gas analysis in oil

Abstract

The present invention belongs to the technical field of transformer fault diagnosis, and in particular, relates to a transformer fault diagnosis method and system based on dissolved gas analysis in oil; the method includes obtaining dissolved gas data and external influencing factor data in transformer oil; constructing a time-space diagram based on dissolved gas data; inputting the time-space diagram and external influencing factors into a trained gas content prediction model, fully modeling the time-space correlation characteristics of time series data in the time dimension and space dimension, and further integrating the external influencing factor data to obtain the gas content prediction value of the future time step. The gas content prediction value is processed by the improved three-ratio method to obtain the fault diagnosis result of the transformer; the present invention models the online monitoring data of dissolved gas in oil into a time-space diagram, realizes full use of the potential connection between the data, can obtain a more accurate gas content prediction value, and realizes accurate online fault diagnosis of the transformer.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer fault diagnosis, and in particular relates to a transformer fault diagnosis method and system based on analysis of dissolved gas in oil.

[0002] Dissolved Gas Analysis (DGA) is a monitoring technology that analyzes the composition and content of dissolved gas in transformer insulating oil. By studying the correlation between dissolved gas in transformer oil and transformer status, the operating status of the transformer can be effectively diagnosed and potential faults can be eliminated in a timely manner.

[0003] Offline testing refers to various tests and inspections after the electrical equipment is shut down. Offline analysis of dissolved gas in transformer oil cannot understand the internal insulation condition of the transformer in real time due to complex operation, long test cycle, and large errors caused by human influence. It is difficult to find latent faults in the equipment as early as possible. Online monitoring can overcome the shortcomings of traditional methods, block the development and expansion of latent faults of transformers as early as possible, and eliminate faults in the bud.

[0004] Transformers are affected by many factors during operation and often suffer from various faults. One of the main manifestations of a fault is the generation of fault characteristic gas. For example, when the transformer overheats, discharges, or the insulation becomes damp, the gas content in the transformer oil will increase rapidly. Therefore, by using the historical gas online monitoring data of the transformer to predict the development trend in the future, the development trend and degree of the transformer's latent faults can be predicted.

[0005] However, most of the current methods for predicting dissolved gas in oil only consider the temporal correlation characteristics of gas content data; however, time series data usually have complex temporal and spatial correlations, and each variable depends not only on its historical value, but also on the influence of other variables and external factors of the transformer. Therefore, the existing methods fail to fully extract the complex correlations between different time series data, and cannot effectively utilize the potential spatial correlation between variables. Therefore, there are relatively large limitations, and it is impossible to accurately diagnose the potential faults of the transformer and the degree of faults at future times. Background Art

[0006] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art. Summary of the invention

[0007] In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides a transformer fault diagnosis method and system based on analysis of dissolved gas in oil.

[0008] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0009] A first aspect of the present invention provides a transformer fault diagnosis method based on dissolved gas analysis in oil, comprising:

[0010] Obtain data on dissolved gas in transformer oil and external influencing factors and perform preprocessing;

[0011] Constructing a spatiotemporal map based on the preprocessed dissolved gas data;

[0012] The space-time graph and external influencing factors are input into the trained gas content prediction model to obtain the predicted value of gas content in the future time step;

[0013] The improved three-ratio method is used to process the predicted value of gas content and obtain the diagnosis result of transformer fault.

[0014] The gas content prediction model includes multiple stacked spatiotemporal feature extraction modules, a fusion layer and an output layer; the stacked spatiotemporal feature extraction modules are used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data; the fusion layer is used to weightedly fuse the spatiotemporal features with the external influencing factor data to obtain the spatiotemporal fusion features that integrate the external influencing factors; the spatiotemporal fusion features pass through the output layer to obtain the predicted value of the gas content in the future time step.

[0015] A second aspect of the present invention provides a transformer fault diagnosis system based on dissolved gas analysis in oil, comprising:

[0016] The data acquisition module is configured to: acquire the data of dissolved gas in transformer oil and external influencing factors and perform preprocessing;

[0017] The space-time graph construction module is configured to: construct a space-time graph based on the preprocessed dissolved gas data;

[0018] The gas content prediction module is configured to: input the spatiotemporal graph and external influencing factors into the trained gas content prediction model to obtain the predicted value of the gas content at the future time step;

[0019] The gas content prediction model includes a plurality of stacked spatiotemporal feature extraction modules, a fusion layer and an output layer; the stacked spatiotemporal feature extraction modules are used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data; the fusion layer is used to weightedly fuse the spatiotemporal features with the external influencing factor data to obtain the spatiotemporal fusion features that fuse the external influencing factors; the spatiotemporal fusion features pass through the output layer to obtain the predicted value of the gas content in the future time step;

[0020] The transformer fault diagnosis module is configured to: process the gas content prediction value using the improved three-ratio method to obtain the transformer fault diagnosis result.

[0021] A third aspect of the present invention provides a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the steps of a transformer fault diagnosis method based on dissolved gas analysis in oil as described in the first aspect of the present invention.

[0022] The fourth aspect of the present invention provides an electronic device, including a memory, a processor, and a program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in a transformer fault diagnosis method based on dissolved gas analysis in oil as described in the first aspect of the present invention are implemented.

[0023] One or more of the above technical solutions have the following beneficial effects:

[0024] (1) The present invention models the online monitoring data of dissolved gas in oil as a space-time graph, and fully models the space-time correlation characteristics of time series data in the time dimension and space dimension through the space-time feature extraction module; and further considers the influence of external factors of the transformer on the gas content in oil, and obtains the space-time fusion characteristics of gas data fusion external factors by weighted fusion of external factors with space-time characteristics, so as to make full use of the potential connection between the data and predict the gas content data for a period of time in the future, so as to obtain a more accurate gas content prediction value and realize accurate online fault diagnosis of the transformer.

[0025] (2) The present invention uses the temporal convolution layer to extract the temporal features of the data. However, the scale of the convolution kernel extraction of the temporal convolution layer is limited. Therefore, in order to fully extract the long-term and short-term dependencies of the time series, a global attention mechanism is introduced to reconstruct the input feature sequence. The reconstructed sequence is then input into the temporal convolution layer to extract the long-term dependency time features, which can fully mine the short-term volatility characteristics and long-term dependency characteristics of the data.

[0026] (3) The present invention comprehensively utilizes a deep learning network and an improved three-ratio method to accurately diagnose the potential faults of the transformer. First, the gas content prediction model is used to obtain the predicted value of the gas content of the dissolved gas in the transformer oil in the future time step. Then, the improved three-ratio method is used to characterize the relationship between the gas content prediction value and the transformer fault, and the nature, type and degree of the transformer fault are judged. This solves the problem that it is difficult to make an accurate judgment on the potential faults of the transformer by relying on a single diagnostic method.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 This is a flow chart of a transformer fault diagnosis method based on dissolved gas analysis in oil according to the first embodiment.

[0030] Figure 2 This is a gas content prediction flow chart of the first embodiment.

[0031] Figure 3 This is a structural diagram of the gas content prediction model of the first embodiment.

[0032] Figure 4 This is a structural diagram of the temporal convolutional layer based on the global attention mechanism of the first embodiment. DETAILED DESCRIPTION

[0033] Embodiment 1

[0034] like Figure 1 As shown, this embodiment discloses a transformer fault diagnosis method based on dissolved gas analysis in oil, comprising:

[0035] Step 1: Obtain the data of dissolved gas in transformer oil and external influencing factors and perform preprocessing;

[0036] Step 2: Construct a spatiotemporal graph based on the preprocessed gas data;

[0037] Step 3: Input the space-time graph and external influencing factor data into the trained gas content prediction model to obtain the predicted value of gas content in the future time step;

[0038] Step 4: Use the improved three-ratio method to process the gas content prediction value to obtain the transformer fault diagnosis result.

[0039] In step 1, when the transformer is in operation, the transformer oil and solid insulating materials will age and deteriorate over time, and will also decompose into gas; by predicting the concentration of dissolved gas in the oil, transformer failures can be monitored and prevented in a timely manner, thereby improving the reliability of the transformer.

[0040] Step 101: Acquire dissolved gas data in transformer oil, including concentration data of hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4 and acetylene C2H2;

[0041] Obtain data on external factors affecting dissolved gas in transformer oil, including transformer oil temperature and load data.

[0042] Transformer insulating oil is based on paraffin (CnH 2n＋2 ) is the main chemical component. During the long-term operation of the transformer, due to arcing, discharge, overheating, moisture and other reasons, the chemical bonds are gradually broken, and fault-related fault discrimination gases (CH4, C2H2, C2H4, C2H6, CO and CO2) are generated.

[0043] Although multiple components of gas may be dissolved in the transformer insulating oil, the contents of CO and CO2 are often much greater than those of hydrogen and hydrocarbon gases, and their changes are not obvious for different fault types and may even be confused with each other; therefore, the invention mainly detects the content of hydrocarbon compounds (hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4 and acetylene C2H2) produced by the breaking of chemical bonds in the insulating oil.

[0044] The gas in transformer oil mainly comes from electrical or thermal faults. During the operation of the transformer, changes in load will cause changes in the internal temperature of the transformer, increase the transformer oil-paper insulation temperature, and further affect the solubility and accumulation of dissolved gas in the transformer oil. This embodiment comprehensively considers the gas production principle of the transformer characteristic gas, as well as the influence of the transformer oil temperature and load on the characteristic gas, and uses the gas concentration data of m nodes at t historical moments and related factors such as oil temperature and load as inputs to the gas content prediction model to predict the gas concentration of the m nodes at future moments (such as the t+Tth moment).

[0045] Step 102: pre-process the acquired data, including supplementing missing values, deleting outliers, and normalizing.

[0046] Among them, the linear interpolation method is used to fill in the missing values ​​of historical data; the singular values ​​are deleted from the completed historical data according to the 3δ rule; and the minimum-maximum normalization method is used to normalize the historical data.

[0047] Data completion includes: assuming that the missing data changes linearly during the missing period, and the two end points of the missing data are determined by known data, constructing a straight line between the two known data points, and obtaining the completion value of the missing data based on the linear law of the two end points; linear interpolation is simple and easy, and is suitable for situations where data is missing in the short term.

[0048] Deleting singular values ​​according to the 3δ rule includes: calculating the mean and standard deviation of the data, where the mean represents the average level of the data and the standard deviation represents the degree of dispersion of the data; determining the normal distribution parameters of the difference between two consecutive measurement points in the completed historical online monitoring data; when the error between a data point and the sample mean exceeds 3 times the standard deviation, the data point is an outlier; by eliminating outliers, the interference with data analysis can be reduced and the accuracy of statistical results can be improved.

[0049] In the historical online monitoring data set, the numerical differences between different features are huge due to dimensional differences. Therefore, the minimum-maximum normalization method is used to normalize the original data so that all feature values ​​are standardized to the [0,1] interval.

[0050] Among them, the minimum-maximum normalization method is used to normalize the dissolved gas data in transformer oil. Specifically, in the historical concentration monitoring sequence, each variable , after normalization, we get the new gas concentration time series data ;

[0051] ;

[0052] In the formula, min Indicates the minimum value in the gas concentration sequence, max Indicates the maximum value in the gas concentration series.

[0053] Normalize the oil temperature of the transformer, including:

[0054] ;

[0055] In the formula, Indicates the oil temperature at the i-th moment, min Indicates the minimum value in the oil temperature series, max Indicates the maximum value in the oil temperature series.

[0056] According to the ratio of the transformer's real-time operating load to the rated load, the transformer's load data is normalized, including:

[0057] ;

[0058] In the formula, represents the real-time load data at the i-th moment, Indicates rated load.

[0059] In step 2, a spatiotemporal graph is constructed based on the preprocessed gas data, including:

[0060] Given a historical monitoring data set from time 0 to T ,in Represents the number of one-dimensional data records, N represents the number of features, and constructs a spatiotemporal graph ;

[0061] In the formula, represents the i-th space-time graph; is the historical time length of the historical online monitoring dataset D, which determines the number of spatiotemporal graphs that can be constructed; M represents the number of nodes contained in a spatiotemporal graph; N represents the number of features contained in a node;

[0062] Space-time diagram for:

[0063]

[0064] Where V represents the node of the space-time graph, and each type of gas data is regarded as a graph node. , X represents the features contained in each node, ; The number of behavioral features of the matrix is ​​N, and the number of columns is M, so the dimension of the matrix is ​​[M,N].

[0065] The spatiotemporal graph G (Graph, G) is a graph data with spatiotemporal information. Since the spatiotemporal graph G has both the data characteristics of the graph and the characteristics of time series data, the particularity of this data requires that the two dimensions of space and time be fully considered during the learning and training process to complete the extraction of spatiotemporal features. The present invention converts the one-dimensional gas content data into two-dimensional graph data by defining an adjacency matrix, so that the model can consider the spatiotemporal correlation of multivariate time series data from the perspective of the graph.

[0066] The correlation between the gas concentration data was calculated based on the Pearson correlation coefficient, and the correlation coefficient between the factors was used as the adjacency matrix A (Adjacency Matrix, A) that describes the connection relationship between nodes in the space-time graph to determine the strength of the correlation between the factors.

[0067] The Pearson correlation coefficient calculation formula is:

[0068]

[0069] in, is the correlation coefficient between gas concentration data X and Y, , are the means of the concentration data of gas X and gas Y respectively.

[0070] In step 3, if Figure 2 , Figure 3 As shown, the sequence prediction model includes a stacked spatiotemporal feature extraction module, a fusion layer, and an output layer; the stacked spatiotemporal feature extraction module is used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data;

[0071] The fusion layer is used to perform weighted fusion of spatiotemporal features with external influencing factor data to obtain spatiotemporal fusion features that integrate external influencing factors; the spatiotemporal fusion features pass through the output layer to obtain the predicted gas concentration value for the future time step.

[0072] Among them, the time dependency of data and the potential correlation between data are mined through stacked spatiotemporal feature extraction modules. The stacked spatiotemporal feature extraction modules include multiple spatiotemporal feature extraction modules connected in sequence, and the output of the previous spatiotemporal feature extraction module is the input of the next spatiotemporal feature extraction module.

[0073] like Figure 4 As shown, the spatiotemporal feature extraction module includes: a temporal convolution layer based on a global attention mechanism, a graph convolution layer, and a gating mechanism;

[0074] The input sequence is input into the temporal convolution layer based on the global attention mechanism. The global attention layer is first used to extract the temporal dependency features of the input sequence at different time scales. To prevent information loss, the output of the global attention layer is concatenated with the initial input sequence and then input into the temporal convolution layer to obtain the temporal features of short-time scale mutation and long-time scale dependency.

[0075] The temporal features output by the temporal convolution layer are input into the graph convolution layer to extract the spatial features between the input sequences. Finally, the temporal features are fused with the spatial features using a gating mechanism to obtain the spatiotemporal features of the input gas sequence.

[0076] Among them, the temporal convolution layer based on the attention mechanism includes: a global attention layer, a temporal convolution layer and a residual connection layer; the global attention layer is used to map the input features of the node into three feature matrices, and after multiple attention branches perform feature extraction at different scales, they are fused using the splicing layer and the linear transformation layer to obtain the reconstructed input features; the temporal convolution layer is used to extract the long-term dependency features of the reconstructed input features, and the output of the temporal convolution layer is added to the initial input features through the residual connection layer to obtain the temporal dependency features.

[0077] The training process of the gas content prediction model includes: based on the spatiotemporal graph data ( arrive The training set is constructed by using the sequential set of all spatiotemporal graphs of the model and the data of external influencing factors; the model is trained multiple times using the training set, and the prediction model is optimized using the Adam algorithm. The trained prediction model is then used to make predictions based on the test set data; the prediction of the model is evaluated based on the prediction results, and the prediction model with the best evaluation index is selected as the trained prediction model.

[0078] Specifically: the error between the predicted value and the true value is calculated by the mean square error function, the prediction model is repeatedly trained, and the new prediction value is obtained by the repeatedly trained prediction model according to the above steps until the error between the new predicted value and the true value is minimized, thereby obtaining the trained final gas content prediction model.

[0079] During the training process of the gas content prediction model, space-time graph data is needed to simultaneously learn spatial and temporal patterns; each input of the model is a sequential set of p space-time graphs and the external influencing factor data corresponding to each space-time graph, where p is a hyperparameter in the model, indicating how many space-time graphs are input each time to train the model; the predicted value output by the model is also two-dimensional data, and its dimension is the same as that of the input space-time graph.

[0080] Step 3 specifically includes:

[0081] Step 301: Time-dependent feature extraction

[0082] Step 3011, extracting time-dependent features using a global attention layer;

[0083] Since the size of the convolution kernel is fixed, the traditional temporal convolution layer can only extract temporal features within the local time step range, but cannot effectively extract global temporal correlation features outside the convolution kernel. Therefore, the present invention introduces a global attention mechanism before the temporal convolution layer, so that the model can capture global temporal dependency features across time steps without restriction.

[0084] Specifically: The global attention mechanism takes the input sequence Linear mapping to query matrix Q , key matrix K , value matrix V .

[0085]

[0086] In the formula, , , are the learnable weight matrices for the query matrix, key matrix, and value matrix respectively;

[0087] Split the current input sequence into multiple attention heads in the feature dimension r , obtain the feature information of different subspaces; multiple attention branches are finally spliced ​​to obtain the global time-dependent features;

[0088]

[0089]

[0090] In the formula, represents the attention weight, Indicates Attention branches;

[0091] Finally, in order to avoid information loss, the global time-dependent feature With the input features Add channel by channel to get the final feature tensor , as the input of the temporal convolutional layer.

[0092] Step 3012, the temporal convolution layer includes two stacked dilated convolution layers (i.e., the first dilated convolution layer and the second dilated convolution layer), a residual connection layer, and a 1x1 convolution layer. The multi-scale temporal features are extracted by using the two stacked dilated convolution layers. The input features are added to the output of the dilated convolution by using the residual connection layer, and then the number of channels is adjusted by the 1x1 convolution layer to obtain the final temporal features.

[0093] Specifically: The input of the temporal convolutional layer is After two layers of dilated convolution operations, the second dilated convolution is performed to obtain the feature tensor Feature tensor convolved with the first dilated And the input feature tensor The output after addition is realized to map the input sequence into features containing time information; by introducing the residual connection layer, the gradient vanishing problem caused by the deep model network structure is eliminated.

[0094] Among them, the hole convolution operation at the tth time step is:

[0095]

[0096] In the formula, represents the dilated convolution, d is the dilation factor that controls the dilation size, and by adjusting d during training, we can focus on features of different distances. is the input vector, represents the convolution function, Represents the size of the convolution kernel in the convolution function, represents the learnable weight matrix; Indicates the past direction The input vector.

[0097] No. The output of the temporal convolutional layer is:

[0098]

[0099] in, is the number of time steps; is the temporal feature output by the Lth temporal convolutional layer; Γ represents the convolution operation module consisting of hole convolution, nonlinear activation function and weight normalization. is a learnable weight vector, Represents the vector addition operation used to implement residual connections.

[0100] Step 302: Spatial correlation feature extraction

[0101] The observation values ​​of each node at different times in the space-time graph have obvious correlation with the observation values ​​of its neighboring multiple time steps. In the time dimension, the mutual influence between the historical data and future data of the same node is mined through the time convolution layer; in the spatial dimension, the feature sequence output by the time convolution layer is input into the graph convolutional neural network layer GCN (Graph Convolutional Network, GCN) to mine the mutual influence between different nodes at the same time;

[0102] From the perspective of the space-time graph, multivariate time series data can be regarded as nodes on the graph. The attributes of each node are represented by the feature matrix, and the interconnected relationships between nodes are described by the adjacency matrix A of the space-time graph; the adjacency matrix A of the space-time graph is the correlation weight matrix between the gas data calculated in step 2.

[0103] No. The output of the graph convolutional neural network layer is:

[0104]

[0105] In the formula, represents the spatial features of the Lth layer, represents the temporal features of the Lth layer, represents the activation function, represents element-wise multiplication, A represents the adjacency matrix of the space-time graph, represents the learnable weight matrix;

[0106] Step 303: Fusion of temporal and spatial features

[0107] The outputs of the temporal convolution layer and graph convolution layer in the L-th layer feature extraction module are and , and finally through the gating mechanism and Fusion to obtain spatiotemporal features ;

[0108] ;

[0109] ;

[0110] In the formula, , , represents the learnable weight matrix, express Activation function, H represents the hidden state.

[0111] Step 304: Integrate external influencing factors

[0112] The present invention considers two types of external influencing factors, namely oil temperature and load data; the values ​​of oil temperature and load are normalized to the interval [0,1];

[0113] The external influencing factor fusion layer includes a 1x1 convolutional layer and a fully connected layer connected in sequence, which concatenates the preprocessed oil temperature and load data to obtain the external factor vector , input into the 1x1 convolutional layer, adjust the dimension of the external influencing factor data, and obtain the external factor feature matrix with the same spatiotemporal feature dimension as the spatiotemporal feature extraction module output ;

[0114] Output of 1x1 convolutional layer The output of the spatiotemporal feature extraction module Then, the data is weighted and fused through a fully connected layer to obtain the spatiotemporal fusion features of the external factors. .

[0115] Step 305: Output layer

[0116] The output layer includes a Dropout layer and a fully connected layer connected in sequence; the Dropout layer is used to prevent overfitting of the network by modifying the number of neurons in the hidden layer during training; the fully connected layer is used to map the input spatiotemporal fusion features to the output space.

[0117] Integrating the spatiotemporal fusion characteristics of external factors After passing through the output layer, the gas concentration prediction results for the future time period are obtained.

[0118] In step 4, the gas content prediction value is processed using the improved three-ratio method, including:

[0119] The improved three-ratio method is an effective method currently widely used in transformer fault diagnosis. The improved three-ratio method calculates the three ratios of C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6, and determines the nature of the transformer fault by looking up the table according to the known coding rules and classification methods; among them, transformer faults include: low-temperature overheating fault, medium-temperature overheating fault, high-temperature overheating fault, partial discharge fault, low-energy discharge fault, and arc discharge fault.

[0120] The coding rules and fault type judgment method of the improved three-ratio method are shown in Table 1, and the coding combinations of the improved three-ratio method and their corresponding transformer fault types are shown in Table 2.

[0121] Table 1 Coding of the improved three-ratio method

[0122]

[0123] Table 2 Transformer fault types

[0124]

[0125] To summarize, this embodiment takes into account the complexity of the transformer failure mechanism and the diversity of fault signs. Therefore, it first uses the gas content prediction model constructed based on the deep learning network to predict the dissolved gas content in the oil for a period of time in the future, and then uses the improved three-ratio method to judge the nature, category and degree of the potential fault of the transformer, thereby solving the problem that it is difficult to make an accurate judgment on the potential fault of the transformer by relying on a single diagnostic method.

[0126] Embodiment 2

[0127] This embodiment discloses a transformer fault diagnosis system based on dissolved gas analysis in oil, comprising:

[0128] The data acquisition module is configured to: acquire the data of dissolved gas in transformer oil and external influencing factors and perform preprocessing;

[0129] The space-time graph construction module is configured to: construct a space-time graph based on the preprocessed dissolved gas data;

[0130] The gas content prediction module is configured to: input the spatiotemporal graph and external influencing factors into the trained gas content prediction model to obtain the predicted value of the gas content at the future time step;

[0131] The gas content prediction model includes a plurality of stacked spatiotemporal feature extraction modules, a fusion layer and an output layer; the stacked spatiotemporal feature extraction modules are used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data; the fusion layer is used to weightedly fuse the spatiotemporal features with the external influencing factor data to obtain the spatiotemporal fusion features that fuse the external influencing factors; the spatiotemporal fusion features pass through the output layer to obtain the predicted value of the gas content in the future time step;

[0132] The transformer fault diagnosis module is configured to: process the gas content prediction value using the improved three-ratio method to obtain the transformer fault diagnosis result.

[0133] Embodiment 3

[0134] The purpose of this embodiment is to provide a computer-readable storage medium.

[0135] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in a transformer fault diagnosis method based on dissolved gas analysis in oil as described in Example 1 of the present disclosure.

[0136] Embodiment 4

[0137] The purpose of this embodiment is to provide an electronic device.

[0138] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, the steps in a transformer fault diagnosis method based on dissolved gas analysis in oil as described in Example 1 of the present disclosure are implemented.

[0139] The steps involved in the apparatuses of the above embodiments 2, 3 and 4 correspond to the method embodiment 1, and the specific implementation methods can refer to the relevant description part of embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0140] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0141] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. A transformer fault diagnosis method based on dissolved gas analysis in oil, characterized in that: include: Obtain data on dissolved gas in transformer oil and external influencing factors and perform preprocessing; Constructing a spatiotemporal map based on the preprocessed dissolved gas data; The space-time graph and external influencing factors are input into the trained gas content prediction model to obtain the predicted value of gas content in the future time step; The improved three-ratio method is used to process the predicted value of gas content to obtain the diagnosis result of transformer fault; the improved three-ratio method is to calculate the three ratios of C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6, and determine the nature of transformer fault by looking up the table according to the known coding rules and classification methods; The gas content prediction model includes a plurality of stacked spatiotemporal feature extraction modules, a fusion layer, and an output layer; the stacked spatiotemporal feature extraction modules are used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data; The fusion layer is used to perform weighted fusion of the spatiotemporal features with the external influencing factor data to obtain the spatiotemporal fusion features that integrate the external influencing factors; the spatiotemporal fusion features pass through the output layer to obtain the predicted value of the gas content in the future time step.

2. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 1, characterized in that: The dissolved gas data in the transformer oil include: concentration data of hydrogen H2, methane CH4, ethane C2H6, ethylene C2H4 and acetylene C2H2; The external influencing factor data of the gas dissolved in the transformer oil include: oil temperature and load data of the transformer.

3. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 1, characterized in that: The acquired data are preprocessed, including supplementing missing values, deleting outliers, and normalizing.

4. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 3, characterized in that: The missing values ​​of historical data are filled by linear interpolation method; singular values ​​are deleted from the filled historical data according to the 3δ rule; the historical data are normalized by minimum-maximum normalization method.

5. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 1, characterized in that: The spatiotemporal feature extraction module includes a temporal convolution layer, a graph convolution layer and a gating mechanism based on an attention mechanism; The time convolution layer based on the attention mechanism is used to extract multi-scale time-dependent features of nodes in the spatiotemporal graph; The features output by the temporal convolution layer are input into the graph convolution layer to extract the spatial correlation features between nodes in the spatiotemporal graph; The gating mechanism is used to fuse the outputs of the temporal convolution layer and the graph convolution layer to obtain the spatiotemporal features.

6. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 5, characterized in that: The temporal convolution layer based on the attention mechanism includes: a global attention layer, a temporal convolution layer and a residual connection layer; the global attention layer is used to map the input features of the node into three feature matrices, and after the features of different scales are extracted by multiple attention branches, they are fused by a splicing layer and a linear change layer to obtain reconstructed input features; the temporal convolution layer is used to extract the long-term dependency features of the reconstructed input features, and the output of the temporal convolution layer is added to the initial input features through the residual connection layer to obtain the temporal dependency features.

7. A transformer fault diagnosis method based on dissolved gas analysis in oil as claimed in claim 1, characterized in that: The transformer faults include: low temperature overheating fault, medium temperature overheating fault, high temperature overheating fault, partial discharge fault, low energy discharge fault and arc discharge fault.

8. A transformer fault diagnosis system based on dissolved gas analysis in oil, characterized in that: include: The data acquisition module is configured to: acquire the data of dissolved gas in transformer oil and external influencing factors and perform preprocessing; The space-time graph construction module is configured to: construct a space-time graph based on the preprocessed dissolved gas data; The gas content prediction module is configured to: input the spatiotemporal graph and external influencing factors into the trained gas content prediction model to obtain the predicted value of the gas content at the future time step; The gas content prediction model includes a plurality of stacked spatiotemporal feature extraction modules, a fusion layer, and an output layer; the stacked spatiotemporal feature extraction modules are used to extract features in the time dimension and the space dimension to obtain the spatiotemporal features of the gas data; The fusion layer is used to perform weighted fusion of the spatiotemporal features and the external influencing factor data to obtain the spatiotemporal fusion features that integrate the external influencing factors; the spatiotemporal fusion features are passed through the output layer to obtain the predicted value of the gas content in the future time step; The transformer fault diagnosis module is configured to: use the improved three-ratio method to process the gas content prediction value to obtain the diagnosis result of the transformer fault; the improved three-ratio method is to calculate the three ratios of C2H2 / C2H4, CH4 / H2, and C2H4 / C2H6, and determine the nature of the transformer fault by looking up the table according to the known coding rules and classification methods.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the steps of a transformer fault diagnosis method based on dissolved gas analysis in oil as described in any one of claims 1 to 7 are implemented.

10. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the transformer fault diagnosis method based on dissolved gas analysis in oil as described in any one of claims 1 to 7 are implemented.

Images