Transformer state evaluation method, device and equipment and storage medium

Abstract

The invention provides a transformer state evaluation method, which comprises the steps of selecting characteristic parameters influencing the state of a transformer, calculating weights by adopting subjective and objective weighting methods, determining a combined weight based on a game theory model, and obtaining operation state change data by combining the combined weight and the characteristic parameters. And inputting a pre-constructed future running state prediction model of the transformer, and evaluating the state of the transformer. According to the method, the transformer operation data is analyzed to obtain the characteristic parameter weight relation, the important characteristic parameters are extracted, the state synthesis method is provided, the transformer degradation degree is accurately evaluated, and the future operation situation is predicted.

Description

Technical Field

[0001] This application belongs to the field of transformer status prediction, and particularly relates to a transformer status evaluation method, device, equipment, and storage medium. Background Art

[0002] Power transformers play a core role in the power system, connecting different voltage levels. Their safe and reliable operation is crucial for the safety of the power grid. Although the traditional regular maintenance method can extend the service life of equipment based on experience, due to differences in the environment and operating conditions, it often leads to insufficient or excessive maintenance, resulting in waste of resources. In fact, research shows that only 6% of faults are related to time, and regular maintenance only evaluates from the time dimension, so its effectiveness is limited. With the increasing complexity and reliability requirements of the power grid, as well as the difficulties in outage arrangements and the limited maintenance resources, the limitations of regular maintenance have become increasingly obvious. Summary of the Invention

[0003] Based on the technical problems stored in the above background art, this application proposes a transformer status evaluation method, device, equipment, and storage medium.

[0004] A transformer status evaluation method proposed by this application includes:

[0005] Select characteristic parameters that affect the transformer status;

[0006] Calculate three groups of weights of the characteristic parameters respectively using a subjective weighting method and two objective weighting methods;

[0007] Determine the combined weight of the characteristic parameters among the three groups of weights based on the game theory model;

[0008] Combine the combined weight and the characteristic parameters to obtain the operation status change data of the transformer;

[0009] Input the operation status change data into a pre-constructed transformer future operation status prediction model to evaluate the status of the transformer.

[0010] Optionally, the subjective weighting method includes the expert weighting method, i.e., the G1 method; the objective weighting methods include the entropy weight method and the CRITIC method.

[0011] Optionally, determining the combined weight based on the game theory model includes:

[0012] Define the three groups of weights of the characteristic parameters as a weight vector set;

[0013] For any linear combination of k vectors in the weight vector set, obtain the combined weight U, and the calculation formula of U is:

[0014]

[0015] Among them, U represents the combined weight, and α k is the weight coefficient;

[0016] Optimize the combined weight, and the expression is as follows;

[0017]

[0018] According to the differential properties of the matrix, obtain the optimal first-order derivative condition of the above formula, and the expression is as follows:

[0019]

[0020] According to the first-order derivative condition, obtain the symmetric model, and the expression is as follows:

[0021]

[0022] According to the symmetric model, obtain the combined coefficients of each weight, and the expression is as follows:

[0023]

[0024] Obtain the combined weight according to the combined coefficients, and the expression is as follows:

[0025]

[0026] Optionally, the prediction model for the future operating state of the transformer includes: a CNN part and a GRU part;

[0027] The CNN part includes an input layer, a convolutional layer, a pooling layer, and a fully connected layer;

[0028] The GRU part includes an input layer, a GRU layer, and a fully connected layer;

[0029] The output of the CNN part is used as the input of the GRU part.

[0030] Optionally, after constructing the combined neural network, it further includes: selecting the hyperparameters of the combined neural network through Bayesian optimization.

[0031] Optionally, after selecting the hyperparameters of the combined neural network through Bayesian optimization, it further includes: correcting the credibility coefficients of each characteristic parameter of the transformer.

[0032] Optionally, the characteristic parameters at least include:

[0033] Oil chromatography, temperature, and electrical quantities, including hydrogen content, oil temperature, and core grounding current.

[0034] The present application also provides a transformer status evaluation device, including:

[0035] A selection module, configured to select characteristic parameters affecting the transformer status;

[0036] A weight module, configured to calculate three groups of weights of the characteristic parameters respectively by using a subjective weighting method and two objective weighting methods;

[0037] A game module, configured to determine the combined weight of the characteristic parameters among the three groups of weights based on a game theory model;

[0038] A merging module, configured to combine the combined weight and the characteristic parameters to obtain the operation status change data of the transformer;

[0039] A pre-charging module, configured to input the operation status change data into a pre-constructed transformer future operation status prediction model to evaluate the transformer status.

[0040] Optionally, the subjective weighting method includes the expert weighting method, i.e., the G1 method; the objective weighting methods include the entropy weight method and the CRITIC method.

[0041] Optionally, determining the combined weight based on the game theory model includes:

[0042] Defining the three groups of weights of the characteristic parameters as a weight vector set;

[0043] Performing an arbitrary linear combination on k vectors in the weight vector set to obtain the combined weight U, and the calculation formula of U is:

[0044]

[0045] where U represents the combined weight, and α k is the weight coefficient;

[0046] Optimizing the combined weight, and the expression is as follows;

[0047]

[0048] According to the differential property of the matrix, obtaining the optimal first-order derivative condition of the above formula, and the expression is as follows:

[0049]

[0050] According to the first-order derivative condition, obtaining a symmetric model, and the expression is as follows:

[0051]

[0052] According to the symmetric model, obtaining the combined coefficients of each weight, and the expression is as follows:

[0053]

[0054] The combination weight is obtained according to the combination coefficient, and the expression is as follows:

[0055]

[0056] Optionally, the transformer future operating state prediction model includes: a CNN part and a GRU part;

[0057] The CNN part includes input layer, convolution layer, pooling layer and fully connected layer;

[0058] The GRU part includes an input layer, a GRU layer and a fully connected layer;

[0059] The output of the CNN part serves as the input of the GRU part.

[0060] Optionally, after constructing the combined neural network, the method further includes: selecting hyperparameters of the combined neural network through Bayesian optimization.

[0061] Optionally, after selecting the hyperparameters of the combined neural network through Bayesian optimization, the method further includes: correcting the credibility coefficients of each characteristic parameter of the transformer.

[0062] Optionally, the characteristic parameters include at least:

[0063] Oil chromatography, temperature and electrical quantities, including hydrogen content, oil temperature, core ground current.

[0064] The present application also provides a transformer status assessment device, comprising:

[0065] A memory for storing a computer executable program of the transformer condition assessment method;

[0066] The processor is configured to call a computer executable program in the memory to execute the following steps: selecting characteristic parameters that affect a transformer state; calculating three sets of weights for the characteristic parameters using a subjective weighting method and two objective weighting methods respectively; determining a combined weight of the characteristic parameters in the three sets of weights based on a game theory model; obtaining transformer operating state change data by combining the combined weight and the characteristic parameters; and inputting the operating state change data into a pre-built transformer future operating state prediction model to evaluate the transformer state.

[0067] The present application also provides a storage medium storing a computer executable program, which is used to be called by a processor to execute the steps of the above-mentioned transformer state assessment method.

[0068] Advantages and beneficial effects of this application:

[0069] This application proposes a transformer state assessment method, comprising: selecting characteristic parameters that affect the transformer state; calculating three sets of weights for the characteristic parameters using a subjective weighting method and two objective weighting methods; determining the combined weights of the characteristic parameters in the three sets of weights based on a game theory model; combining the combined weights and the characteristic parameters to obtain transformer operating state change data; and inputting the operating state change data into a pre-built transformer future operating state prediction model to assess the transformer state. This application analyzes transformer operating data to obtain the characteristic parameter weight relationships that affect the operating status, extracts important characteristic parameters, and proposes a state synthesis method to accurately assess the degree of transformer degradation and predict its future operating status. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 This is a schematic diagram of the transformer condition assessment process in this application;

[0071] Figure 2 This is a schematic diagram of the transformer status assessment device in this application. DETAILED DESCRIPTION

[0072] The present application will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present application and implement it.

[0073] The following contents are all examples of specific implementation processes provided for detailed description of the technical solutions to be protected by this application. However, this application is also implemented in other ways different from the descriptions herein. Under the guidance of the concept of this application, those skilled in the art may implement this application using different technical means. Therefore, this application is not limited to the specific embodiments below.

[0074] Please refer to Figure 1 As shown, a transformer status assessment method comprises the following steps:

[0075] S101 selects characteristic parameters that affect the transformer state.

[0076] Before conducting a transformer condition assessment, it is crucial to select appropriate transformer characteristic parameters to ensure the reliability of the assessment results. Numerous characteristic parameters related to transformer condition exist, which can be broadly categorized as online monitoring data and experimental data. Based on the concept of situational awareness, this application selects characteristic parameters for real-time online monitoring to assess the transformer's operating status, enabling real-time monitoring and prediction of the transformer's future condition.

[0077] Based on the above principles, the following three characteristic parameters were selected as primary parameters for transformer condition assessment: oil color spectrum, temperature, and electrical quantity. Furthermore, to more comprehensively assess the transformer's condition, 10 secondary characteristic parameters were also selected, including hydrogen content in transformer oil, oil temperature, and core grounding current. The specific classification is shown in Table 1:

[0078] Table 1 Division of transformer characteristic parameters

[0079]

[0080] These characteristic parameters were selected based on their close correlation with the transformer's condition. Oil chromatography analysis reveals potential faults and aging processes within the transformer; temperature monitoring reflects the transformer's thermal condition and prevents overheating; and electrical quantity measurements reveal the transformer's electrical performance and insulation condition.

[0081] S102 calculates three sets of weights of the characteristic parameters using a subjective weighting method and two objective weighting methods respectively.

[0082] When building a transformer condition assessment model, calculating the weights of characteristic parameters is crucial and directly impacts the accuracy of the assessment results. Traditional transformer condition assessment models often only consider the information content, variability, and correlation of characteristic parameter data when calculating weights, and simply use a weighted approach to combine subjective and objective weights. The weights calculated by these methods can differ significantly from actual conditions, potentially causing the assessment results to misalign with the actual operating status of the transformer.

[0083] In order to overcome the above problems, this application adopts a subjective and objective combined weighting method to calculate the weights of the characteristic parameters. Specifically, the subjective weighting method adopts the G1 method. The above method avoids the shortcomings of the hierarchical analysis method and has strong operability and practicality. The objective weighting rule adopts two methods: the entropy weight method and the CRITIC method. The entropy weight method is a method that can determine the weight by the amount of information contained in the data. The smaller the entropy value of the characteristic parameter, the greater the information contained, and the higher the corresponding weight. The CRITIC rule calculates the objective weight based on the contrast intensity of the characteristic parameters and the correlation between the parameters. It takes into account the variability of the characteristic parameters and the correlation, making the evaluation more scientific and objective.

[0084] Determination of subjective weight:

[0085] When determining the subjective weight, we mainly rely on the experience and knowledge of experts to rank the importance of each secondary feature parameter and calculate the corresponding weight. The specific steps are as follows:

[0086] First, based on the experts' experience, the importance of each secondary characteristic parameter was ranked. This ranking is based on the experts' in-depth understanding and practical experience of transformer condition assessment, and comprehensively considers the role and influence of each characteristic parameter on transformer operation.

[0087] After the ranking is complete, the importance ratios of adjacent feature parameters need to be assigned. This ratio reflects the relative difference in importance between adjacent feature parameters. This assignment is still based on the experience and knowledge of experts, who will comprehensively consider the impact and scope of the feature parameters to determine a relatively reasonable importance ratio.

[0088] After assigning the importance ratio, the weight w of the mth feature parameter is calculated according to the formula m , the expression is as follows:

[0089]

[0090] The above weights reflect the importance and influence of the characteristic parameter in transformer condition assessment. During calculation, a recursive method is used to gradually derive the weights of subsequent characteristic parameters based on the known ratio of importance and the weight of the previous characteristic parameter.

[0091] After calculating the weight of the mth feature parameter, the weights of other feature parameters are derived using the ratio of the above weight to the known importance level. The expression is as follows:

[0092]

[0093] The above process is a reverse derivation process, starting from the feature parameter with known weight, and gradually deriving the weights of other feature parameters. In the derived formula, w k-1 is the weight of the k-1th feature parameter, r k is the ratio of importance, w k is the weight of the w-th feature parameter.

[0094] Through the above steps, the subjective weights of each secondary characteristic parameter are determined. These weights reflect the experts' views on the importance and influence of each characteristic parameter in transformer condition assessment. These subjective weights will serve as an important reference for the subsequent subjective and objective combined weighting method to calculate the final combined weights.

[0095] Determination of entropy weight method weights:

[0096] The entropy weight method is a method to determine the weight based on the amount of data information. ij (i=1,2,…,n;j=1,2,…,m) is the jth data under the i-th characteristic parameter. For the given data, if x ijThe greater the difference, the greater its information entropy, which means that the feature parameter contains more information.

[0097] When calculating the weights of the entropy weight method, it is necessary to calculate the information entropy of each feature parameter first. Let e j be the entropy weight of the j-th feature parameter, and its calculation formula is:

[0098]

[0099] Among them, e j > 0, p ij is the proportion of the j-th data of the i-th feature parameter in all data of the i-th feature parameter, which can be calculated by the following formula:

[0100]

[0101] Here, k is a constant used to ensure 0 ≤ e j ≤ 1. In practical applications, the value of k is usually 1 / ln(n), which can make the value range of e j between 0 and 1.

[0102] Finally, a set of objective weights can be obtained through the entropy weight method. This set of weights reflects the importance and influence of each feature parameter in the transformer state assessment. These objective weights will be used as an important reference basis for the subsequent subjective and objective combined weighting method to calculate the final combined weights. By integrating subjective and objective information, a more accurate and reliable transformer state assessment result can be obtained.

[0103] Calculation of weights by CRITIC method:

[0104] The CRITIC method is a method for determining objective weights based on the contrast intensity and the conflict between indicators. The above method comprehensively considers the value gap size of the evaluation schemes within different indicators and the correlation between indicators to calculate the objective weights of each indicator.

[0105] First of all, it is necessary to calculate the standard deviation s j of the j-th feature parameter. The above standard deviation reflects the dispersion degree of the values of each evaluation scheme within the feature parameter. The greater the standard deviation, the greater the value gap of the feature parameter and the higher the contrast intensity. The calculation formula of the standard deviation is as follows:

[0106]

[0107] Next, it is necessary to calculate the conflict cj between the j-th feature parameter and other parameters. The above conflict reflects the correlation between the feature parameter and other feature parameters. The lower the correlation, the higher the conflict. The calculation formula of the conflict is as follows:

[0108]

[0109] where r ij is the correlation coefficient between the characteristic parameters i and j. The value range of the correlation coefficient is between -1 and 1. 1 indicates a perfect positive correlation, -1 indicates a perfect negative correlation, and 0 indicates no correlation. Therefore, 1 - r ij reflects the conflict between the characteristic parameters i and j.

[0110] Finally, calculate the information quantity Cj of the j-th characteristic parameter. The above information quantity comprehensively considers two factors: the comparison intensity and the conflict. Its calculation formula is as follows:

[0111] c j = s j R j ;

[0112] The larger Cj is, the greater the information quantity contained in the j-th characteristic parameter, and the greater the relative importance of this index. Therefore, the objective weight wj of the j-th index can be calculated by the following formula:

[0113]

[0114] Through the above method, the weights of each characteristic parameter can be objectively determined according to the comparison intensity of the data and the conflict between the indexes. These objective weights will be used as an important reference for the subsequent subjective and objective combined weighting method to calculate the final combined weight. By integrating subjective and objective information, a more accurate and reliable transformer status evaluation result can be obtained. At the same time, the weight calculation process of the CRITIC method also reflects the in-depth mining and analysis of data, which helps to discover potential information and rules in the data.

[0115] S103 determines the combined weight of the characteristic parameters in the three groups of weights based on the game theory model.

[0116] Game theory is the theory and method for studying competitive things, and it is an important discipline in the field of operations research. Game theory analyzes the rational behavior and decision-making equilibrium of multiple decision-making subjects when their behaviors interact. In game theory, it can be assumed that each scheme is the result of rational decision-making, which is a decision made by the decision-maker to maximize its own interests or minimize its own losses. This competitive result is not controlled by a single decision-maker, but is jointly achieved by all decision-makers.

[0117] The purpose of game theory combined weighting is to optimize and combine the weights calculated by various methods to obtain the optimal weight value.

[0118] In this application, three methods are used to assign weights to the characteristic parameters respectively to obtain three groups of characteristic parameter weight vectors. Game theory is used to find the weight combination that minimizes the deviation between the weight combination and each weight.

[0119] For a basic weight vector set U i (i = 1, 2, …, n), any linear combination of these n vectors forms a possible combined weight U:

[0120]

[0121] where U represents the possible combined weight, and α k is the weight coefficient.

[0122] The method of game theory can be used to find the most suitable weight combination in the set of possible vectors. Its basic idea is to find agreement or compromise among different weights. Finding the most satisfactory weight vector can be transformed into optimizing the linear combination weight coefficient wi, and the optimization goal is to minimize the deviation between the weight combination and each weight. That is:

[0123]

[0124] According to the differential properties of the matrix, the optimal first derivative condition of the above formula can be obtained as:

[0125]

[0126] According to the differential properties of the matrix, the optimal first derivative condition of the above formula can be obtained as:

[0127] wi = (Wi^TWi)^-1Wi^TWi. Thus, the countermeasure model is derived as follows:

[0128]

[0129] After obtaining α1, …, α L , normalization processing is performed to obtain the combined coefficients of each weight:

[0130]

[0131] Finally, the combined weight is obtained:

[0132]

[0133] S104 combines the combined weight and the characteristic parameters to obtain the operating state change data of the transformer.

[0134] After obtaining the combined weight, the operating state change data of the transformer can be obtained by combining the combined weight and the transformer state parameter data.

[0135] S105 inputs the operation state change data into a pre - constructed prediction model for the future operation state of the transformer to evaluate the state of the transformer.

[0136] First, construct a prediction model for the future operation state of the transformer based on BO - CNN - GRU. This future operation state prediction model includes a CNN part and a GRU part.

[0137] The CNN neural network is a feed - forward neural network widely used in the field of deep learning. It extracts a high - dimensional feature matrix to generate feature vectors, thereby retaining feature information while reducing the data dimension and the risk of overfitting of the prediction model.

[0138] CNN mainly consists of an input layer, a convolutional layer, a pooling layer, a fully - connected layer, and an output layer. The input feature vector can be a multi - dimensional vector group, and it adopts the methods of local perception and weight sharing. The convolutional layer extracts feature quantities from the original data and deeply explores the internal relationships of the data. The pooling layer can reduce the network complexity and the number of training parameters. The fully - connected layer merges the processed data and calculates the classification and regression results.

[0139] CNN is stacked by multiple convolutional layers and multiple pooling layers. A single - layer CNN network consists of 2 layers: 1 convolutional layer and 1 pooling layer, which can directly process the original input sequence. Each layer of CNN contains several convolutional kernels of the same size and the same type of pooling function. First, the convolutional kernel traverses the entire input sequence data to generate a higher - level and more abstract feature space; second, the pooling layer compresses each generated feature for secondary feature extraction and dimensionality reduction, and selects the important features at a higher level; finally, new sequence features are generated as the input for the next convolutional layer and pooling layer.

[0140] The main steps of CNN are as follows:

[0141] Convolutional layer: The convolutional neural network takes the output of the upper layer as the input of the lower layer, and the core is the convolutional layer. Convolutional calculation realizes local perception and weight sharing through feature mapping. The convolutional kernel window size slides to select part of the data for calculation, and the result of convolution is the feature map. Usually, a convolutional layer has multiple convolutional kernels, which will generate multiple feature maps, and the weights of the same convolutional kernel are shared. This feature reduces the number of network connections, reduces the model complexity, and reduces the system memory expenditure.

[0142] Pooling layer: Used to reduce the number of parameters and prevent overfitting, generally using max - pooling.

[0143] Fully - connected layer: Reduces the network dimension and outputs the feature vector.

[0144] In practical applications, multiple layers of convolution are used, followed by fully connected layers for training. The purpose of using multiple layers of convolution is to reduce the number of network parameters and learning data. At the same time, the features learned by a single layer of convolution are often local. The higher the number of layers, the more global the features learned.

[0145] RNNs are suitable for analyzing and processing time series data because they introduce a recurrent unit structure into the network and allow for internal connections between hidden units, making it possible to explore the temporal relationships between discontinuous data. However, RNNs suffer from the vanishing gradient problem, which causes them to lose the ability to learn information from the distant past as the time interval increases.

[0146] The introduction of LSTM neural network solves the problem of RNN gradient disappearance and has been widely used in the field of time series data prediction. In recent years, many variants have evolved according to different needs.

[0147] GRU is an improved model of LSTM. It integrates the forget gate and input gate into a single update gate and mixes the neuron state and hidden state. It can effectively alleviate the "gradient vanishing" problem in recurrent neural networks and reduce training parameters while maintaining training effects. At the same time, GRU can better capture the dependencies of sequence time steps, taking into account the correlation and nonlinear relationship of time series data, and realize the prediction of the future operating status of the transformer.

[0148] In this application, the update gate z in GRU t The larger the value, the more state information from the previous moment is brought into the current state, and the reset gate r t The smaller it is, the more information of the previous state is written into the current candidate set h t The reset gate r in GRU t and update gate z t The current time step input x t and the hidden state h of the previous time step t-1 , calculated in the fully connected layer of the sigmoid activation function, the calculation formula is:

[0149] z t =σ(x t w xz +h t-1 w hz +b z );

[0150] r t =σ(x t w xr +h t-1 w hr +b r );

[0151] In this application, the candidate hidden state h in the GRU t is obtained by element-wise multiplying the output of the reset gate r at the current time step t with the hidden state h at the previous time step t-1 Then, in the fully connected layer with the tanh activation function, it is calculated by connecting the input x at the current time step t with the output of the element-wise multiplication. The calculation formula is as follows:

[0152]

[0153] The hidden state h in the GRU t is calculated by using the update gate Z t to combine the hidden state h at the previous time step t-1 and the candidate hidden state at the current time step The formula is as follows:

[0154]

[0155] In summary, the process of the GRU neural network can be simplified as:

[0156] h t = GRU(x t , h t-1 );

[0157] By extending information through multiple GRU units, long-term sequence prediction is achieved.

[0158] Furthermore, the Bayesian optimization method can be considered the most advanced optimization framework currently. It can be applied to many fields, not limited to hyperparameter search, but also to advanced fields such as neural network architecture search and meta-learning.

[0159] In this application, for the complex optimization problem of unknown objective function expressions and high search costs of hyperparameters in deep learning models, a continuously updated probability model is used in Bayesian optimization. By evaluating the objective function a few times, the posterior probability of the optimization function is updated to obtain the optimal combination of model hyperparameters. The selection of the hyperparameter combination of the model can be expressed as:

[0160] x * = argmin f(x);

[0161] where f(x) is the minimized objective function used to evaluate the optimal performance of the objective function; x * is the finally obtained optimal hyperparameter combination.

[0162] Bayesian optimization originates from Bayes' theorem and uses the BO formula to establish the probability distribution of the optimization process:

[0163] P(E|D) ∝ P(D|E)P(E);

[0164] Where P(E) is a Gaussian distribution; P(E|D) is a Gaussian regression process, which can be determined by the kernel matrix ∑. ∑ is defined by a kernel function, and its expression is:

[0165]

[0166] The purpose of the BO hyperparameters is to obtain an optimal combination of hyperparameters through continuous iteration, that is, x * = x n+1 to make the accuracy of the model reach the optimal.

[0167] Finally, in the process of predicting the state of the transformer, this application not only needs to consider the numerical values of the characteristic parameters, but also needs to consider their credibility. Since different characteristic parameters may have different credibility, this application needs to correct them.

[0168] In practical applications, this application can correct the credibility coefficients of each characteristic parameter of the transformer through the following steps:

[0169] Calculate the credibility coefficient of each characteristic parameter. The credibility coefficient can be obtained through methods such as statistical analysis of historical data, expert experience, or experimental tests.

[0170] Correct the credibility coefficient of each characteristic parameter. This application can use the following method for correction: multiply the credibility coefficient by the weight of the characteristic parameter to obtain the corrected weight.

[0171] Apply the corrected weight to the training and prediction processes of the BO-CNN-GRU combined neural network. During training and prediction, this application not only needs to consider the numerical values of each characteristic parameter, but also needs to consider their credibility. Therefore, the corrected weight needs to be introduced into the calculation of the neural network.

[0172] Repeat the above steps until a satisfactory prediction result is achieved. By correcting the credibility coefficients of each characteristic parameter of the transformer, this application can further improve the accuracy and robustness of the state prediction of the BO-CNN-GRU combined neural network.

[0173] This application also provides a transformer state evaluation device, including:

[0174] A selection module, used to select the characteristic parameters that affect the state of the transformer;

[0175] A weight module, used to calculate three groups of weights of the characteristic parameters respectively by using a subjective weighting method and two objective weighting methods;

[0176] A game module, configured to determine the combined weight of the characteristic parameters in the three sets of weights based on a game theory model;

[0177] A merging module, configured to combine the combined weight and the characteristic parameters to obtain the operation state change data of the transformer;

[0178] A pre-charging module, configured to input the operation state change data into a pre-constructed prediction model for the future operation state of the transformer to evaluate the state of the transformer.

[0179] This application also provides a transformer state evaluation device, including:

[0180] A memory, configured to store the computer executable program of the above-mentioned transformer state evaluation method;

[0181] A processor, configured to call the computer executable program in the memory and execute: selecting characteristic parameters affecting the state of the transformer; respectively calculating three sets of weights of the characteristic parameters by using a subjective weight assignment method and two objective weight assignment methods; determining the combined weight of the characteristic parameters in the three sets of weights based on a game theory model; combining the combined weight and the characteristic parameters to obtain the operation state change data of the transformer; inputting the operation state change data into a pre-constructed prediction model for the future operation state of the transformer to evaluate the state of the transformer.

[0182] This application also provides a storage medium, storing a computer executable program, which is used to be called by a processor to execute the steps of the above-mentioned transformer state evaluation method.

Claims

1. A transformer status evaluation method, characterized in that, Including: Selecting characteristic parameters that affect the state of the transformer; Calculating three groups of weights of the characteristic parameters respectively by using a subjective weighting method and two objective weighting methods; Determining the combined weight of the characteristic parameters in the three groups of weights based on the game theory model; Combining the combined weight and the characteristic parameters to obtain the operation state change data of the transformer; Inputting the operation state change data into a pre-constructed prediction model of the future operation state of the transformer to evaluate the state of the transformer.

2. The transformer status evaluation method according to claim 1, characterized in that The subjective weighting method includes the expert weighting method, i.e., the G1 method; the objective weighting methods include the entropy weight method and the CRITIC method.

3. The transformer status evaluation method according to claim 1, wherein Determining the combined weight based on the game theory model includes: Defining the three groups of weights of the characteristic parameters as a weight vector set; Performing an arbitrary linear combination on k vectors in the weight vector set to obtain the combined weight U, and the calculation formula of U is: Among them, U represents the combined weight, and α k is the weight coefficient; Optimizing the combined weight, and the expression is as follows; According to the differential property of the matrix, obtaining the optimal first derivative condition of the above formula, and the expression is as follows: According to the first derivative condition, obtaining a symmetric model, and the expression is as follows: According to the symmetric model, obtaining the combined coefficients of each weight, and the expression is as follows: Obtaining the combined weight according to the combined coefficients, and the expression is as follows:

4. The transformer status evaluation method according to claim 1, characterized in that The prediction model of the future operation state of the transformer includes: a CNN part and a GRU part; The CNN part includes an input layer, a convolutional layer, a pooling layer and a fully connected layer; The GRU part includes an input layer, a GRU layer and a fully connected layer; The output of the CNN part is used as the input of the GRU part.

5. The transformer status evaluation method according to claim 4, wherein After constructing the combined neural network, it further includes: selecting the hyperparameters of the combined neural network through Bayesian optimization.

6. The transformer status evaluation method according to claim 5, wherein After selecting the hyperparameters of the combined neural network through Bayesian optimization, it further includes: correcting the credibility coefficients of each characteristic parameter of the transformer.

7. The transformer status evaluation method according to claim 1, characterized in that The characteristic parameters at least include: Oil chromatography, temperature and electrical quantities, including hydrogen content, oil temperature, and core grounding current.

8. A transformer condition assessment device, characterized in that, Including: A selection module for selecting characteristic parameters that affect the state of the transformer; A weight module for calculating three groups of weights of the characteristic parameters respectively by using a subjective weighting method and two objective weighting methods; A game module for determining the combined weight of the characteristic parameters in the three groups of weights based on the game theory model; A merging module for combining the combined weight and the characteristic parameters to obtain the operation state change data of the transformer; A pre-charging module for inputting the operation state change data into a pre-constructed prediction model of the future operation state of the transformer to evaluate the state of the transformer.

9. A transformer condition assessment device, characterized in that, Including: A memory for storing the computer executable program of the transformer state evaluation method according to any one of claims 1 to 7; A processor for calling the computer executable program in the memory to execute: selecting characteristic parameters that affect the state of the transformer; Three groups of weights of the characteristic parameters are calculated by using a subjective weight assignment method and two objective weight assignment methods respectively; the combined weight of the characteristic parameters in the three groups of weights is determined based on a game theory model; the operation state change data of the transformer is obtained by combining the combined weight and the characteristic parameters; the operation state change data is input into a pre-constructed prediction model for the future operation state of the transformer to evaluate the state of the transformer.

10. A storage medium, characterized in that, A computer executable program is stored, and the computer executable program is used to be retrieved by a processor to execute the steps of the transformer state evaluation method according to any one of claims 1 to 7.

Images