Top oil temperature prediction method and system based on transformer

Abstract

The invention relates to the technical field of transformer top oil temperature prediction, in particular to a top oil temperature prediction method and system based on a transformer. Acquiring historical monitoring data of the transformer, wherein the historical monitoring data comprise top oil temperature, winding temperature, high-voltage side active power, reactive power, phase voltage and phase current data; a K-nearest neighbor interpolation method is adopted to process missing values, and an isolated forest algorithm is adopted to eliminate abnormal points; an improved energy valley optimization algorithm IEVO is used for optimization, and the algorithm adopts a polynomial variation disturbance strategy and combines three decay modes of alpha, beta and gamma to update parameters; extracting data frequency components through VMD decomposition, performing nonlinear dimensionality reduction on top oil temperature data in combination with Kernel PCA, and constructing an SHAP-MLP feature selection model to screen important features; and finally, based on the SOFTS model, the model processes time sequence data through an STAR module, a multi-layer perceptron is adopted to extract features, and residual connection is used to enhance the expression ability. According to the invention, high-precision prediction of the top oil temperature of the transformer is realized, and a reliable basis can be provided for equipment maintenance.

Description

Technical Field

[0001] The invention relates to the technical field of transformer top oil temperature prediction, and in particular to a transformer-based transformer top oil temperature prediction method and system. Background Art

[0002] With the continuous growth of electricity demand and the increasing requirements for the reliability of power equipment, the safe operation and performance monitoring of transformers as key equipment in the power system are particularly important. The change of the top oil temperature of the transformer directly affects the working efficiency and service life of the equipment. Traditional temperature monitoring methods mostly rely on fixed monitoring points and simple linear models, which are difficult to reflect the dynamic changes of oil temperature in a timely manner, leading to potential equipment failures and energy waste. In recent years, with the advancement of data acquisition technology and the development of artificial intelligence, prediction methods based on big data and intelligent algorithms have gradually become an effective means to solve this problem. Variational mode decomposition (VMD) can effectively extract useful information from data, kernel principal component analysis (Kernel PAC) helps with dimensionality reduction and denoising, SHAP feature selection can identify key features, and the SOFTS model is widely used in performance prediction of various complex systems with its superior time series prediction capabilities.

[0003] Based on this background, the present invention proposes a transformer top oil temperature prediction method and system based on the VMD-Kernel PAC-SHAP-SOFTS model, which realizes high-precision prediction of transformer top oil temperature by comprehensively utilizing the above advanced technologies. This method can not only reflect temperature changes in a timely manner, but also provide a scientific basis for transformer maintenance and management, and has important practical application value and good market prospects. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is how to solve the problem that traditional transformer temperature monitoring methods mostly rely on fixed monitoring points and simple linear models, which are difficult to reflect the dynamic changes of oil temperature in a timely manner, leading to potential equipment failures and energy waste.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a top oil temperature prediction method based on a transformer, which includes obtaining historical monitoring data of the transformer, the historical monitoring data including top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and preprocessing the historical monitoring data;

[0008] The improved energy valley algorithm is used to optimize the parameters of the SOFTS deep learning model that inputs preprocessed historical monitoring data;

[0009] The optimized top oil temperature data is subjected to VMD decomposition and Kernel PCA dimensionality reduction, and feature selection is performed on the transformer top oil temperature features.

[0010] Based on the data after the feature selection and dimensionality reduction processing, a prediction model is established and the top oil temperature of the transformer is predicted.

[0011] As a preferred solution of the transformer-based top oil temperature prediction method of the present invention, the preprocessing of the historical monitoring data includes: processing the missing values ​​in the historical monitoring data using the K nearest neighbor interpolation method;

[0012] The isolation forest algorithm is used to detect and remove abnormal points in the historical monitoring data.

[0013] As a preferred solution of the transformer-based top oil temperature prediction method described in the present invention, when the K nearest neighbor interpolation method is used for processing, the data is first subjected to Z-core standardization processing; the distance relationship between data points is calculated by Euclidean distance; and the missing values ​​are interpolated by weighted average method.

[0014] As a preferred solution of the transformer-based top oil temperature prediction method described in the present invention, the improved energy valley optimization algorithm includes: optimizing the initial population using a polynomial mutation perturbation strategy; performing α decay, β decay and γ decay updates based on the enrichment level and stability of the particles; wherein the polynomial mutation perturbation strategy includes performing mutation operations and boundary processing on decision variables.

[0015] As a preferred solution of the transformer-based top oil temperature prediction method described in the present invention, the feature selection and dimensionality reduction processing includes: decomposing the top oil temperature data using variational mode decomposition VMD to extract different frequency components; performing nonlinear dimensionality reduction on the decomposed mode function through Kernel PCA; constructing a SHAP-MLP feature selection model to evaluate the importance of transformer top oil temperature features, and selecting features with SHAP values ​​greater than 1.

[0016] As a preferred solution of the transformer-based top oil temperature prediction method described in the present invention, the SHAP-MLP feature selection model includes: an input layer receives the preprocessed transformer top oil temperature features; a hidden layer performs feature conversion through a ReLU activation function; and an output layer generates prediction results and feature importance evaluation.

[0017] As a preferred solution of the transformer-based top oil temperature prediction method described in the present invention, the establishment of the prediction model includes: optimizing the SOFTS model parameters using an improved energy valley optimization algorithm, including the embedding dimension, core representation dimension, number of encoder layers, prediction window length and learning rate; using the top oil temperature data after VMD decomposition and Kernel PCA dimensionality reduction and the features after SHAP-MLP feature selection as input; processing time series data through the STAR module to capture long-term and short-term dependencies; using a multi-layer perceptron to extract core feature representation; and using encoder layers and residual connections to enhance the model's expressiveness.

[0018] In a second aspect, an embodiment of the present invention provides a top oil temperature prediction system based on a transformer, which includes an acquisition module, which acquires historical monitoring data of the transformer, wherein the historical monitoring data includes top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and pre-processes the historical monitoring data;

[0019] The optimization module uses the improved energy valley algorithm to optimize the parameters of the SOFTS deep learning model that inputs preprocessed historical monitoring data;

[0020] The decomposition and dimensionality reduction and feature selection module performs VMD decomposition and Kernel PCA dimensionality reduction on the optimized top oil temperature data, and performs feature selection on the transformer top oil temperature features. The prediction module establishes a prediction model and predicts the transformer top oil temperature based on the data after the feature selection and top oil temperature decomposition and dimensionality reduction processing.

[0021] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the transformer-based top oil temperature prediction method are implemented as described in the first aspect of the present invention.

[0022] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the transformer-based top oil temperature prediction method are implemented as described in the first aspect of the present invention.

[0023] The beneficial effects of the present invention are as follows: the present invention pre-processes historical monitoring data by introducing the K nearest neighbor interpolation method and the isolation forest algorithm, wherein the K nearest neighbor interpolation method uses a weighted average method to process missing values, and the isolation forest algorithm can effectively identify and eliminate outliers, thereby solving the problem of reduced prediction accuracy caused by poor data quality in traditional methods, making data pre-processing more intelligent and efficient, and providing a high-quality data foundation for subsequent modeling.

[0024] Through the improved energy valley optimization algorithm (IEVO), the polynomial mutation perturbation strategy is adopted to optimize the initial population, and the three decay modes of α, β, and γ are combined to update the parameters. Compared with the traditional optimization algorithm, it has stronger global search ability and local fine adjustment ability, can converge to the optimal solution faster, and significantly improves the optimization efficiency of the model.

[0025] Through the combined application of VMD decomposition and Kernel PCA dimensionality reduction:

[0026] VMD decomposition can decompose the top oil temperature data into different frequency components and effectively extract data features; KernelPCA performs nonlinear dimensionality reduction, which reduces the data dimension while retaining key information; this combination not only ensures the integrity of the top oil temperature data characteristics, but also improves the computational efficiency.

[0027] Innovatively construct the SHAP-MLP feature selection model:

[0028] The importance of features is evaluated by SHAP value; features with SHAP value greater than 1 are selected as input; this method realizes automatic screening of features, avoids the subjectivity of manual experience selection, and improves the interpretability and prediction accuracy of the model.

[0029] The SOFTS model structure is used to process time series data through the STAR module, combined with multi-layer perceptron and residual connection:

[0030] The STAR module can effectively capture the long-term and short-term dependencies of data; the residual connection enhances the expressiveness of the model; the multi-layer perceptron extracts deep features; this deep learning architecture significantly improves the model's ability to fit nonlinear relationships, making the prediction results more accurate and reliable.

[0031] Synergy of the overall solution:

[0032] Through the complete technical chain of data preprocessing → feature engineering → model optimization → prediction; each link closely cooperates and promotes each other; finally, high-precision prediction of the top oil temperature of the transformer is achieved, and the prediction error is reduced by more than 30% compared with the traditional method, providing reliable technical support for the safe operation and preventive maintenance of the transformer.

[0033] This method can monitor and predict the temperature change trend of the top layer of transformer oil in real time; give early warning of possible temperature anomalies; provide decision-making basis for operation and maintenance personnel; effectively reduce the failure rate of transformer, extend the service life of equipment, and generate significant economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 It is a flow chart of the transformer-based top oil temperature prediction method;

[0036] Figure 2 A diagram of computer equipment for transformer-based top oil temperature prediction method;

[0037] Figure 3 The model framework diagram of the top oil temperature prediction method based on transformer based on SOFTS;

[0038] Figure 4 This is the energy valley algorithm diagram of the transformer-based top oil temperature prediction method using a polynomial variation perturbation strategy optimization;

[0039] Figure 5 This is the framework diagram of the IEVO-SOFTS ventilation control system based on the transformer top oil temperature prediction method. DETAILED DESCRIPTION

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0041] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0042] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0043] Example 1

[0044] Reference Figure 1 to Figure 5 , which is the first embodiment of the present invention, and provides a method for predicting the top oil temperature of a transformer, comprising:

[0045] S100: Obtain historical monitoring data of the transformer, the historical monitoring data including top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and pre-process the historical monitoring data;

[0046] S101: preprocessing the historical monitoring data includes: processing missing values ​​in the historical monitoring data using a K-nearest neighbor interpolation method;

[0047] The isolation forest algorithm is used to detect and remove abnormal points in historical monitoring data.

[0048] S102: When the K nearest neighbor interpolation method is used for processing: first, the data is subjected to Z-core standardization processing; the distance relationship between data points is calculated by Euclidean distance; and the missing values ​​are interpolated by weighted average method.

[0049] S200: Optimizing the pre-processed historical monitoring data using an improved energy valley optimization algorithm;

[0050] S201: The improved energy valley optimization algorithm includes: optimizing the initial population using a polynomial mutation perturbation strategy; updating α decay, β decay and γ decay based on the enrichment level and stability of the particles; wherein the polynomial mutation perturbation strategy includes mutation operations and boundary processing on decision variables.

[0051] S300: performing feature selection and dimensionality reduction processing on the optimized historical monitoring data;

[0052] S301: Feature selection and dimensionality reduction processing include: using variational mode decomposition VMD to decompose the top oil temperature data and extract different frequency components; performing nonlinear dimensionality reduction on the decomposed mode function through Kernel PCA; constructing a SHAP-MLP feature selection model to evaluate the importance of the transformer top oil temperature feature and select features with a SHAP value greater than 1.

[0053] S302: The SHAP-MLP feature selection model includes: the input layer receives the preprocessed historical monitoring data; the hidden layer performs feature conversion through the ReLU activation function; and the output layer generates prediction results and feature importance evaluation.

[0054] S400: Based on the top oil temperature data after feature selection and dimensionality reduction processing, a prediction model is established to predict the top oil temperature of the transformer.

[0055] S401: The establishment of the prediction model includes: using the improved energy valley optimization algorithm to optimize the SOFTS model parameters, including the embedding dimension, core representation dimension, number of encoder layers, prediction window length and learning rate; using the top oil temperature data after VMD decomposition and Kernel PCA dimensionality reduction and the features after SHAP-MLP feature selection as input; processing time series data through the STAR module to capture long-term and short-term dependencies; using a multi-layer perceptron to extract core feature representations; and using encoder layers and residual connections to enhance the model's expressiveness.

[0056] Furthermore, this embodiment also provides a top oil temperature prediction system based on a transformer, comprising:

[0057] An acquisition module is used to acquire historical monitoring data of the transformer, including top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and pre-process the historical monitoring data;

[0058] The optimization module optimizes the pre-processed historical monitoring data using the improved energy valley optimization algorithm;

[0059] A decomposition dimension reduction and feature selection module is used to perform VMD decomposition and Kernel PCA dimension reduction on the optimized top oil temperature data, and perform feature selection on the transformer top oil temperature features;

[0060] The prediction module establishes a prediction model and predicts the top oil temperature of the transformer based on the top oil temperature data after feature selection and dimensionality reduction processing.

[0061] In summary, the present invention preprocesses historical monitoring data by introducing the K-nearest neighbor interpolation method and the isolation forest algorithm. The K-nearest neighbor interpolation method uses weighted averaging to process missing values, and the isolation forest algorithm can effectively identify and eliminate outliers, solving the problem of reduced prediction accuracy caused by poor data quality in traditional methods, making data preprocessing more intelligent and efficient, and providing a high-quality data foundation for subsequent modeling.

[0062] Through the improved energy valley optimization algorithm (IEVO), the polynomial mutation perturbation strategy is adopted to optimize the initial population, and the three decay modes of α, β, and γ are combined to update the parameters. Compared with the traditional optimization algorithm, it has stronger global search ability and local fine adjustment ability, can converge to the optimal solution faster, and significantly improves the optimization efficiency of the model.

[0063] Through the combined application of VMD decomposition and Kernel PCA dimensionality reduction:

[0064] VMD decomposition can decompose the top oil temperature data into different frequency components and effectively extract data features; KernelPCA performs nonlinear dimensionality reduction, which reduces the data dimension while retaining key information; this combination not only ensures the integrity of the top oil temperature data features, but also improves the computational efficiency.

[0065] Innovatively construct the SHAP-MLP feature selection model:

[0066] The importance of features is evaluated by SHAP value; the transformer top layer oil temperature features with SHAP value greater than 1 are selected as input; this method realizes automatic screening of features, avoids the subjectivity of manual experience selection, and improves the interpretability and prediction accuracy of the model.

[0067] The SOFTS model structure is used to process time series data through the STAR module, combined with multi-layer perceptron and residual connection:

[0068] The STAR module can effectively capture the long-term and short-term dependencies of data; the residual connection enhances the expressiveness of the model; the multi-layer perceptron extracts deep features; this deep learning architecture significantly improves the model's ability to fit nonlinear relationships, making the prediction results more accurate and reliable.

[0069] Synergy of the overall solution:

[0070] Through the complete technical chain of data preprocessing → feature engineering → model optimization → prediction; each link closely cooperates and promotes each other; finally, high-precision prediction of the top oil temperature of the transformer is achieved, and the prediction error is reduced by more than 30% compared with the traditional method, providing reliable technical support for the safe operation and preventive maintenance of the transformer.

[0071] This method can monitor and predict the temperature change trend of the top layer of transformer oil in real time; give early warning of possible temperature anomalies; provide decision-making basis for operation and maintenance personnel; effectively reduce the failure rate of transformer, extend the service life of equipment, and generate significant economic benefits.

[0072] Example 2

[0073] Reference Figure 2 - Figure 5 , which is the second embodiment of the present invention, and this embodiment provides a top oil temperature prediction method based on a transformer. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0074] The historical monitoring data of the top oil temperature of the transformer (including winding temperature, top oil temperature, high-voltage side active power, high-voltage side reactive power, high-voltage side phase voltage, high-voltage side phase current, etc.) are collected, and the K-nearest neighbor interpolation method is used to process the missing values ​​of the data, and the isolation forest algorithm is used to detect and eliminate abnormal points in the data.

[0075] Input the collected historical monitoring data of the top layer oil temperature of the transformer (including winding temperature, top layer oil temperature, high-voltage side active power, high-voltage side reactive power, high-voltage side phase voltage, high-voltage side phase current, etc.).

[0076] The K nearest neighbor interpolation method is used to handle missing values ​​of the data. The K neighboring samples that are most similar to the missing value sample are found, and then the characteristic values ​​of these samples are used for interpolation. Before interpolation, the data is Z-core standardized to eliminate the influence of different feature scales:

[0077]

[0078] Where x is the original feature data, x′ is the standardized data, μ is the mean, and σ is the standard deviation. This method converts the data into a form with zero mean and unit standard deviation.

[0079] For each data point with missing values, the Euclidean distance is used to calculate the distance between the data point and other data points. Based on the calculated Euclidean distance value, for each data point with missing values, the K data points closest to it (i.e., nearest neighbors) are found:

[0080]

[0081] Among them, x i and x j are two data points, k represents the kth feature, and n is the total number of features.

[0082] After finding the K nearest neighbors, the missing values ​​are interpolated using a weighted average method using the known values ​​of these neighbors:

[0083]

[0084] x missing is a data point with missing value. i is the value of the i-th neighbor among the K neighbors found. missing ,x i ) is the missing value data point x missin and neighbor x i The distance between is the reciprocal of the distance, As a weighting coefficient, the closer the distance, the greater the weight.

[0085] The feature data interpolated by the K nearest neighbor interpolation method is used to detect and remove abnormal points using the isolation forest algorithm. The isolation forest isolates data points by constructing multiple binary trees (i.e., isolation trees). The construction process of each tree is as follows:

[0086] 1. Randomly select a feature.

[0087] 2. Randomly select a cut point on the feature to divide the data into two subsets.

[0088] 3. Repeat the above steps until each data point is isolated or the maximum tree depth is reached.

[0089] The anomaly score is calculated by the depth of the tree. Anomalies are usually isolated earlier, so when building the tree, the depth of the anomaly is shallow. For each data point x, its anomaly score is calculated based on the length of the isolated path (that is, the depth at which it is isolated in the tree). The formula for the anomaly score s(x,n) is:

[0090]

[0091] h(x) is the path length of data point x in the isolated tree. E(h(x)) is the expected value of the path length of data point x in multiple trees. c(n) is an adjustment term, where n represents the average isolated tree height of the sample number. The specific formula is:

[0092]

[0093] where H(n) is the harmonic number, defined as:

[0094]

[0095] If s(x,n) is close to 1, then x is an outlier, and if s(x,n) is close to 0, then x is a normal point. According to the preset threshold (0.7), the outlier data points with higher scores are removed. According to the anomaly score value of s(x,n), the outliers are detected and removed.

[0096] The energy valley optimization algorithm EVO is improved, and the polynomial mutation perturbation strategy method is used to initialize and optimize the population to obtain the IEVO algorithm.

[0097] Polynomial mutation is a mutation operation method based on evolutionary algorithms. It introduces appropriate mutations in the initialization phase to improve the diversity of solutions, thereby improving the global search capability. Its position after mutation is determined by the following formula:

[0098]

[0099] in Represents a variable The disturbance term is specifically defined as:

[0100]

[0101] Where r is a random number uniformly distributed in the range [0,1], η is a parameter that controls the degree of variation, usually called the variation index. Larger η values ​​lead to smaller disturbances, and smaller η values ​​lead to larger disturbances.

[0102] The polynomial mutation strategy is added to the energy valley optimization algorithm EVO. When the population is initialized, the polynomial mutation perturbation is introduced to optimize the population diversity. First, a set of particles (solution candidates) are randomly initialized in the search space. Each particle represents a potential solution to the optimization problem. Particle X i Represents the current position vector of the i-th particle. The initial position is determined by the following formula:

[0103]

[0104] in, is the jth decision variable of the ith particle. j,min and x j,max are the lower and upper limits of the variable, respectively. rand is a random number uniformly distributed in the range [0,1].

[0105] Based on the generated initial solution, a polynomial mutation perturbation is performed on each particle. Use the polynomial mutation formula:

[0106]

[0107]

[0108] After performing the mutation, if the variable Beyond the upper and lower boundaries [x j,min ,x j,max ], it needs to be pulled back into the valid range:

[0109]

[0110] Determine the particle enrichment limit and determine its neutron enrichment level NEL i , and calculate the enrichment limit EB of the entire particle:

[0111]

[0112] Among them, NEL i is the neutron enrichment level of the ith particle, and n is the total number of particles.

[0113] Calculate the stability level of the particles. According to the objective function value, calculate the stability level SL of each particle. i :

[0114]

[0115] Among them, BS and WS are the global best and worst particle stability levels, corresponding to the minimum and maximum objective function values, respectively.

[0116] In the main search loop, EVO updates the positions of particles based on their enrichment level and stability.

[0117] If the enrichment level of a particle is above the enrichment limit, the particle position is updated according to the alpha (alpha-ray emission) or gamma (gamma-ray emission) decay mode:

[0118] α decay: For particles with higher stability, a new solution is generated Its mathematical representation is:

[0119]

[0120] Among them, X BS is the position of the particle with the best stability.

[0121] γ decay: For particles with higher stability, a new solution is generated The formula is as follows:

[0122]

[0123] Among them, X Ng is with X i The positions of the nearest neighbor particles.

[0124] β decay: If the stability level of a particle is lower than the stability limit SB, β decay occurs, and the particle conducts a large-scale search to generate a new solution. The update formula is:

[0125]

[0126] Among them, X BS is the position of the global optimal particle, X CP is the center position of the particle, r1 and r2 are random numbers used to control the movement of the particle.

[0127] The algorithm stops when the maximum number of iterations or objective function evaluations is reached, and the global optimal solution is selected as the final result.

[0128] VMD is used to decompose the top oil temperature data and extract its different frequency components. Then Kernel PCA is used to perform nonlinear dimensionality reduction on the decomposed modal function (IMF).

[0129] Step 3.1: Input the historical monitoring data x(t) of top oil temperature

[0130] Step 3.2: Set the number of decomposed modes K and the center frequency ω k , the data is decomposed into several intrinsic mode functions (IMFs) by minimizing the following optimization objective:

[0131]

[0132] Among them, u k (t) kth mode function, ω k is the corresponding center frequency. * indicates convolution operation.

[0133] The constraints are: That is, the sum of all modal functions should be able to reconstruct the original data.

[0134] The above constrained optimization problem is solved by the Lagrange multiplier method, and each mode u is optimized. k (t) and the corresponding center frequency ω k .

[0135] Decompose the input data x(t) into K IMF components u k (t) Each component represents a different frequency content of the data.

[0136] Kernel principal component analysis first inputs the IMF components obtained by VMD decomposition.

[0137] Using the Gaussian kernel function k(x i ,x j ) maps the IMF components from the original space to the high-dimensional feature space:

[0138] x→φ(x)

[0139]

[0140] Calculate the kernel matrix K:

[0141] k(x i ,x j )=<φ(x i ),φ(x j )>

[0142] Center the kernel matrix K:

[0143]

[0144] Among them, 1 n is a matrix of all 1s.

[0145] The centralised kernel matrix Perform eigenvalue decomposition and obtain the eigenvalue λ i and the eigenvector α i

[0146]

[0147] Project the data to the direction with the largest eigenvalue to obtain the features after dimensionality reduction:

[0148]

[0149] A SHAP-MLP feature selection model was constructed to evaluate the contribution of each feature to the prediction of top oil temperature, and features with SHAP values ​​greater than 1 were screened out to reduce the input dimension of the model.

[0150] Define the input data, set the input feature matrix X and the target variable y, where: X = [x1, x2, ..., x n ] is a feature matrix consisting of n input features. y is the prediction target of top oil temperature.

[0151] Construct an MLP model. MLP is a feedforward neural network that includes an input layer, a hidden layer, and an output layer. Its basic structure is as follows: the input layer size is n, which is the number of input features. The hidden layer can be set to any number of neurons. Assume that there are h hidden layer neurons. The output layer size is 1, corresponding to the top oil temperature prediction result.

[0152]

[0153] Among them, W1 and W h are the weight matrices from input to hidden layer and from hidden layer to output layer, b1 and b h is the bias term, σ(·) is the Relu activation function, and f(·) is the activation function of the output layer, which is used to predict the top oil temperature. MLP extracts feature information layer by layer by linearly combining and nonlinearly activating the input features, and finally outputs the predicted value

[0154] SHAP value is an interpretation method based on Shapley value, which is used to measure the contribution of each feature to the model output. Given input feature X = [x1, x2, ..., x n ]SHAP value φ i Reflect feature x i Forecast results contribution.

[0155] For model output The total change, SHAP value assigned to each feature and its contribution is:

[0156]

[0157] Among them, φ0 is the baseline output of the model (i.e., the expected contribution of all features), φ iis feature x i The SHAP value indicates the impact of the feature on the model output. The SHAP value allocates the contribution of each feature by considering the impact of all feature combinations on model prediction.

[0158] Features i The SHAP value formula is:

[0159]

[0160] Where S is a subset of the input features. n is the total number of features. Represents feature subset S plus feature x i The model output after is the model output for feature subset S, by considering feature x i The marginal contribution to different subsets S is used to measure x i The average impact on the final forecast.

[0161] Determine the feature contribution threshold, SHAP value greater than 1, as the standard for screening important features. That is:

[0162]

[0163] in, It is a filtered feature set, containing features with SHAP values ​​greater than 1.

[0164] Optimize the SOFTS model parameters including the embedding dimension d through IEVO embed , core representation dimension d core , the number of encoder layers L enc , prediction window length T pred , learning rate η;

[0165] The goal of optimization is to minimize the prediction error (such as mean square error, MSE) of the SOFTS model by adjusting the following key parameters:

[0166]

[0167] A set of candidate solutions (particles) are initialized according to the polynomial mutation perturbation strategy, each of which corresponds to a different model parameter setting. The dimension of each particle in the population is

[0168]

[0169] η i ∈[1e-5,1e-2], the embedding dimensions range from 10 to 100, 1 to 10, 10 to 200, and 1e-5 to 1e-2 respectively.

[0170] The initial value of each particle is based on the formula:

[0171]

[0172] Generate an initial solution and perturb it according to a polynomial mutation strategy:

[0173]

[0174]

[0175] For each particle (i.e. each set of parameter combinations), training is performed on the SOFTS model, and the fitness function of the model is calculated (usually the loss function on the validation set, mean square error MSE). The fitness function calculation formula is:

[0176]

[0177] in: is the predicted value of the SOFTS model trained using the parameters, y true is the true value.

[0178] According to the fitness value of each particle, the enrichment limit EB and neutron enrichment level NEL are calculated i The stability level SL is calculated by the following formula i :

[0179]

[0180] BS is the global best fitness (smallest MSE), and BS is the global worst fitness (largest MSE).

[0181] According to the update mechanism of IEVO, it is divided into α, β and γ decay processes:

[0182] Alpha decay: Perform local search on particles with high enrichment levels to generate new solutions

[0183]

[0184] Gamma decay: Neighborhood search for particles with high enrichment levels to generate new solutions

[0185]

[0186] Beta decay: global search for particles with low stability levels generates new solutions

[0187]

[0188] After each round of update, the fitness value of each particle is re-evaluated and the global optimal solution X is used. BS renew.

[0189] When the model converges, the globally optimal parameter group is selected as the final parameters of the SOFTS model.

[0190] A transformer top oil temperature prediction model based on IEVO-SOFTS is constructed. The top oil temperature data after VMD decomposition and KernelPCA nonlinear dimensionality reduction and the features after SHAP-MLP feature selection, including winding temperature, transmitter active power, high-voltage side active power, high-voltage side phase voltage, high-voltage side phase current, and high-voltage side reactive power, are used as input to predict the transformer top oil temperature.

[0191] The IEVA-SOFTS model is constructed, and the feature representation obtained by VMD decomposition and Kernel PCA dimensionality reduction is:

[0192] X VMD-KPCA =[IMF1,IMF2,...,IMF k ]

[0193] Among them, IMF k It is the Kth mode function after VMD decomposition and the eigenvector obtained after dimensionality reduction by Kernel PCA.

[0194] The features selected by the SHAP-MLP model include:

[0195] X SHAP =[T 绕组 ,P 变送器有功 ,P 高压侧有功 ,V 高压侧相电压 ,I 高压侧相电流 ,P 高压侧无功 ] [1]

[0196] Combining the above features together, the final input feature is expressed as:

[0197] X input =[IMF1,IMF2,...,IMF k ,T 绕组 ,P 变送器有功 ,P 高压侧有功 ,V 高压侧相电压 ,I 高压侧相电流 ,P 高压侧无功 ]

[0198] Step 6.2: The embedding layer embeds the input data into a high-dimensional space for subsequent hierarchical processing:

[0199] S0=Embedding(Xt )

[0200] The goal of the embedding layer is to transform the raw input data into a high-dimensional vector form so that the model can more effectively capture the complex relationships between input features.

[0201] The STAR module is used for time series processing. The STAR module is based on the attention mechanism and can capture the long-term and short-term dependencies of time series data:

[0202] S i =STAR(S i-1 )

[0203] S i represents the feature representation of the output of the i-th layer in the STAR module, S i-1 Represents the output feature representation of the i-1th layer in the STAR module.

[0204] Extract the core representation through a multi-layer perceptron (MLP):

[0205] o i =MLP(S i-1 )

[0206] S i-1 is the feature representation input to the i-th layer MLP, o i It is the core representation of the output.

[0207] Use repetition and concatenation operations to combine feature representations together:

[0208] F i =Repeat_Concat(S i-1 ,o i )

[0209] Among them, F i Represents the combined features of the feature representation input to the i-th layer MLP and the core representation of the output.

[0210] The encoder layer processes the combined feature representation and combines the original features with the processed features through residual connections to enhance the expressiveness of the model.

[0211] S i =MLP(F i )+S i-1

[0212] Finally, the prediction result is output through the linear layer:

[0213]

[0214] in, represents the final transformer top oil temperature prediction result, SL It represents the final feature representation obtained after layer-by-layer processing by a multi-layer neural network.

[0215] Example 3

[0216] This embodiment also provides a computer device, which is applicable to a transformer-based top oil temperature prediction method, including a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement a forced oscillation detection and positioning method for a distribution network as proposed in the above embodiment.

[0217] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, a forced oscillation detection and positioning method for a distribution network is implemented as proposed in the above embodiment.

[0218] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0219] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0220] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0221] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0222] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used to implement: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0223] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the top oil temperature of a transformer, characterized in that: include, Acquire historical monitoring data of the transformer, the historical monitoring data including top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and pre-process the historical monitoring data; Perform VMD decomposition and Kernel PCA dimensionality reduction on the optimized top oil temperature data, and perform feature selection on the transformer top oil temperature features; The improved energy valley algorithm is used to optimize the parameters of the SOFTS deep learning model that inputs preprocessed historical monitoring data; Based on the data after the feature selection and dimensionality reduction processing, a prediction model is established and the top oil temperature of the transformer is predicted.

2. The transformer top oil temperature prediction method according to claim 1, characterized in that: The preprocessing of the historical monitoring data includes: processing the missing values ​​in the historical monitoring data using the K nearest neighbor interpolation method; The isolation forest algorithm is used to detect and remove abnormal points in the historical monitoring data.

3. The transformer top oil temperature prediction method according to claim 2, characterized in that: When the K nearest neighbor interpolation method is used for processing: first, the data is subjected to Z-core standardization processing; the distance relationship between data points is calculated by Euclidean distance; and the missing values ​​are interpolated by using a weighted average method.

4. The transformer top oil temperature prediction method according to claim 3, characterized in that: The improved energy valley optimization algorithm includes: optimizing the initial population using a polynomial mutation perturbation strategy; updating α decay, β decay and γ decay based on the enrichment level and stability of particles; wherein the polynomial mutation perturbation strategy includes mutation operations and boundary processing on decision variables.

5. The transformer top oil temperature prediction method according to claim 4, characterized in that: The feature selection and dimensionality reduction processing includes: using variational mode decomposition (VMD) to decompose the top oil temperature data and extract different frequency components; using Kernel PCA to perform nonlinear dimensionality reduction on the decomposed mode function; constructing a SHAP-MLP feature selection model to evaluate the importance of the transformer top oil temperature feature and select features with a SHAP value greater than 1.

6. The transformer top oil temperature prediction method according to claim 5, characterized in that: The SHAP-MLP feature selection model includes: an input layer receives the preprocessed transformer top oil temperature features; a hidden layer performs feature conversion through a ReLU activation function; and an output layer generates prediction results and feature importance evaluation.

7. The transformer top oil temperature prediction method according to claim 6, characterized in that: The establishment of the prediction model includes: using an improved energy valley optimization algorithm to optimize the SOFTS model parameters, including embedding dimension, core representation dimension, number of encoder layers, prediction window length and learning rate; using the top oil temperature data after VMD decomposition and Kernel PCA dimensionality reduction and the transformer top oil temperature features after SHAP-MLP feature selection as input; processing time series data through the STAR module to capture long-term and short-term dependencies; using a multi-layer perceptron to extract core feature representation; and using encoder layers and residual connections to enhance the model's expressiveness.

8. A transformer-based top oil temperature prediction system, based on the transformer-based top oil temperature prediction method according to any one of claims 1 to 7, characterized in that: Also includes, An acquisition module is used to acquire historical monitoring data of the transformer, wherein the historical monitoring data includes top oil temperature data, winding temperature data, high-voltage side active power data, high-voltage side reactive power data, high-voltage side phase voltage data, and high-voltage side phase current data, and pre-process the historical monitoring data; The optimization module uses the improved energy valley algorithm to optimize the parameters of the SOFTS deep learning model that inputs preprocessed historical monitoring data; The decomposition dimension reduction and feature selection module performs VMD decomposition and KernelPCA dimension reduction on the optimized top oil temperature data, and performs feature selection on the transformer top oil temperature features. The prediction module establishes a prediction model and predicts the oil temperature at the top layer of the transformer based on the data after the feature selection and decomposition and dimensionality reduction processing.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the transformer-based top oil temperature prediction method are implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the transformer-based top oil temperature prediction method are implemented.

Images