Method and system for predicting top oil temperature of transformer

Abstract

The invention relates to the technical field of transformer top oil temperature prediction, and discloses a transformer top oil temperature prediction method and system, and the method comprises the steps: obtaining first data of a target transformer, and carrying out the first preprocessing of the first historical data, and obtaining second data; establishing a first prediction model and a second correction model connected in series with the first prediction model, and recording the connected models as a third model; a first improved algorithm is preset, hyper-parameter optimization is conducted on the third model according to the first improved algorithm, and hyper-parameters comprise regularization factors and the number of hidden layer nodes; and taking the second data as the input of the third model, and predicting the top oil temperature of the target transformer according to the output of the third model. Through the operation, the top oil temperature of the transformer can be accurately predicted under the small sample condition, and a powerful guarantee is provided for safe operation of a power system.

Description

Technical Field

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

[0002] A transformer is a key device in the power system, and its stable operation is crucial for the generation, transmission, distribution, and use of electricity. During operation, a transformer generates heat, and excessive temperature may cause aging or even damage to the insulating material, leading to faults. Therefore, accurate monitoring and prediction of the transformer temperature are of great significance for ensuring the safe operation of the power system.

[0003] Currently, the methods for obtaining the transformer oil temperature mainly include the direct temperature measurement method, the thermal circuit model method, the numerical calculation method, and the intelligent model algorithm. The direct temperature measurement method collects the temperature in real time through a temperature sensor, but has problems such as high cost and complex maintenance, and its application is limited. The thermal circuit model method predicts the oil temperature by analyzing the heat generation and dissipation processes inside the transformer, which has a clear physical meaning, but the model establishment and parameter calculation are complex, relying on a large amount of experimental data, and there is uncertainty. The numerical calculation method uses the finite element or finite volume method to establish a physical model and obtains the temperature field distribution through iterative solution. Although the prediction is accurate, it consumes a large amount of computing resources and is difficult to meet the real-time requirement. The intelligent model algorithm has been widely used in transformer oil temperature prediction due to its flexibility and accuracy. For example, methods based on the U-net neural network, long short-term memory network (LSTM), Kalman filter algorithm, BP neural network, etc. have achieved good prediction effects in different scenarios. However, the transformer oil temperature is affected by the change of the load rate and has great uncertainty. A single model is difficult to cope with the complex historical data distribution, and the prediction accuracy may be limited. To improve the prediction accuracy, data preprocessing technologies such as principal component analysis (PCA), ensemble empirical mode decomposition (EEMD), etc. are particularly important to reduce the data complexity and noise interference. However, in practical applications, the data available for modeling is less and may be interfered by unknown factors, resulting in bias in the training data. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for predicting the top oil temperature of a transformer, which can solve a series of problems existing in the traditional method for predicting the top oil temperature of a transformer. The direct temperature measurement method can provide direct temperature measurement, but has certain limitations in terms of cost, maintenance, response speed, and accuracy. The thermal circuit model method has complex model establishment and parameter calculation and relies on a large amount of data, with relatively large uncertainties. The numerical calculation method consumes a large amount of computing resources and has a strong dependence on empirical formulas, making it difficult to meet the requirements of real-time performance and high applicability. Single intelligent model algorithms are often sensitive to the distribution of historical data and difficult to learn the potential laws of historical data. When the complexity of historical data increases, the prediction accuracy often decreases significantly, etc.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, the present invention provides a method for predicting the top oil temperature of a transformer, including:

[0008] Obtain the first data of the target transformer, and perform a first preprocessing on the first historical data to obtain the second data;

[0009] The first preprocessing includes a first decomposition operation and a first reconstruction operation;

[0010] Establish a first prediction model and a second correction model connected in series with the first prediction model, and denote the series-connected model as the third model;

[0011] Preset a first improvement algorithm, and optimize the hyperparameters of the third model according to the first improvement algorithm. The hyperparameters include a regularization factor and the number of hidden layer nodes;

[0012] Use the second data as the input of the third model, and predict the top oil temperature of the target transformer according to the output of the third model.

[0013] As a preferred solution of the method for predicting the top oil temperature of a transformer according to the present invention, wherein: the first decomposition operation and the first reconstruction operation include:

[0014] The first decomposition operation is used to decompose the first data, and the first data is the oil temperature data of the target transformer;

[0015] The first reconstruction operation is used to reconstruct the first data decomposed by the first decomposition operation;

[0016] The first reconstruction operation extracts features from the first data decomposed by the first decomposition operation by calculating the distribution entropy to obtain the second data.

[0017] As a preferred solution of the transformer top oil temperature prediction method described in the present invention, wherein: the second correction model connected in series with the first prediction model includes:

[0018] Obtain an error sequence, where the error sequence is a difference sequence between the predicted value of the first prediction model and the true oil temperature value of the target transformer;

[0019] Establish a second correction model based on the error sequence, where the input of the second correction model is the error sequence and the output is the error prediction value;

[0020] Connect the second correction model in series to the output position of the first prediction model.

[0021] As a preferred solution of the transformer top oil temperature prediction method described in the present invention, wherein: the preset first improvement algorithm includes improving the differential evolution algorithm by introducing Tent initialization and golden sine optimization strategy to obtain the first improvement algorithm.

[0022] As a preferred solution of the transformer top oil temperature prediction method described in the present invention, wherein: the hyperparameter optimization of the third model according to the first improvement algorithm includes:

[0023] Establish an optimization objective function based on the third model;

[0024] Take the hyperparameters as the population type of the first improvement algorithm;

[0025] Perform hyperparameter optimization on the third model according to the first improvement algorithm and determine whether the termination condition is satisfied;

[0026] Output the hyperparameters that satisfy the termination condition and update the third model.

[0027] As a preferred solution of the transformer top oil temperature prediction method described in the present invention, wherein: the signal decomposition of the first data includes:

[0028] Add positive and negative random Gaussian white noise to the Nth power of the first data;

[0029] Use the EMD algorithm to decompose the first data after adding white noise;

[0030] Take the average value of each IMF component obtained by decomposition as the final decomposition result, and the final decomposition result is in the form of a data matrix.

[0031] As a preferred solution of the transformer top oil temperature prediction method described in the present invention, wherein: the feature extraction of the first data after the first decomposition operation by calculating the distribution entropy includes:

[0032] Calculate the maximum absolute difference between each vector in the final decomposition result and all other vectors;

[0033] Establish a histogram based on the maximum absolute difference and perform a segmentation operation on the histogram;

[0034] Calculate the number of occurrences and probability density of the maximum absolute error in each segmentation region;

[0035] Calculate the distribution entropy according to the probability density, and complete the feature extraction of the first data after the first decomposition operation.

[0036] In a second aspect, the present invention provides a transformer top oil temperature prediction system, including:

[0037] A preprocessing module, configured to obtain first data of a target transformer and perform first preprocessing on the first historical data to obtain second data;

[0038] The first preprocessing includes a first decomposition operation and a first reconstruction operation;

[0039] A model establishment module, configured to establish a first prediction model and a second correction model connected in series with the first prediction model, and record the series-connected model as a third model;

[0040] A model optimization module, configured to preset a first improvement algorithm and optimize the hyperparameters of the third model according to the first improvement algorithm, where the hyperparameters include a regularization factor and the number of hidden layer nodes;

[0041] A model prediction module, configured to use the second data as the input of the third model and predict the top oil temperature of the target transformer according to the output of the third model.

[0042] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and the processor implements the steps of the method described above when executing the computer program.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and the computer program implements the steps of the method described above when executed by a processor.

[0044] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes a method for predicting the top oil temperature of a transformer. By obtaining the first data of the target transformer and performing a first preprocessing on the first historical data, second data is obtained; a first prediction model and a second correction model connected in series with the first prediction model are established, and the series-connected model is denoted as the third model; a first improvement algorithm is preset, and according to the first improvement algorithm, hyperparameters of the third model are optimized, and the hyperparameters include a regularization factor and the number of hidden layer nodes; the second data is used as the input of the third model, and the top oil temperature of the target transformer is predicted according to the output of the third model. Through the above operations, accurate prediction of the top oil temperature of the transformer can be achieved under small sample conditions, providing a strong guarantee for the safe operation of the power system.

[0045] Specifically, the CEEMD algorithm is used to decompose the historical top oil temperature data of the transformer, decomposing the complex data into multiple subsequences, reducing the data complexity, and helping the subsequent model prediction.

[0046] The DistEn data reconstruction method is used to reconstruct the top oil temperature data of the transformer, which can effectively extract the potential features and information in the top oil temperature data of the transformer, providing more accurate and meaningful input data for the subsequent prediction model, thereby improving the performance of the prediction model.

[0047] Aiming at the problems of slow convergence speed and easy to fall into local optimum in the optimization process of the differential evolution algorithm, the Tent initialization and the Gold-Sine optimization strategy (Gold-SA) are added to the initialization stage and the population update process of the differential evolution algorithm to help the algorithm jump out of the local optimum, thereby constructing the TGDE algorithm, enhancing the search ability of the original differential evolution algorithm, and improving the search efficiency.

[0048] Aiming at the problems of high cost, limited response speed, dependence on a large amount of data and poor robustness in the traditional method for predicting the top oil temperature of a transformer, the present invention adopts the ORedRVFL model to effectively simplify the complexity of the model, improving the robustness of the model while improving the efficiency; and uses TGDE to optimize the parameters in the ORedRVFL model, constructing the CEEMD-DistEn-TGDE-ORedRVFL-EC model, improving the performance and accuracy of the top oil temperature prediction model of the transformer. This method can achieve accurate prediction of the top oil temperature of the transformer under small sample conditions, providing a strong guarantee for the safe operation of the power system. Description of the Drawings

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0050] Figure 1 It is a flowchart of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0051] Figure 2 It is a detailed overall step flowchart of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0052] Figure 3 It is a CEEMD decomposition algorithm structure diagram of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0053] Figure 4 It is an ORedRVFL model structure diagram of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0054] Figure 5 It is a TGDE algorithm flowchart of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0055] Figure 6 It is a fitting comparison diagram of the prediction result and the real result of a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention.

[0056] Figure 7 It is an internal structure diagram of a computer device for a method for predicting the top oil temperature of a transformer provided by an embodiment of the present invention. Specific Embodiments

[0057] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will provide a detailed description of the specific embodiments of the present invention with reference to the accompanying drawings of the specification. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0058] Example 1, referring to Figures 1-7 , which is the first embodiment of the present invention. This embodiment provides a method for predicting the top oil temperature of a transformer, including:

[0059] In the existing related technologies, there are some problems. For example, although the direct temperature measurement method can provide direct temperature measurement, it is costly, complex to maintain, and has limited response speed; although the thermal circuit model method has clear physical meaning, the process of establishing its model and calculating parameters is complex, and it depends on a large amount of experimental data, with great uncertainty; although the numerical calculation method has accurate prediction, it consumes a large amount of computing resources and has a strong dependence on empirical formulas, making it difficult to meet the requirements of real-time and high applicability; while a single intelligent model algorithm is often sensitive to the distribution of historical data and difficult to learn the potential laws of historical data. When the complexity of historical data increases, the prediction accuracy often drops significantly.

[0060] This application provides a method that can effectively solve the above-mentioned problems. Next, multiple embodiments will be combined to elaborate in detail on how to implement the transformer top oil temperature prediction method;

[0061] Figure 1 The method flow chart of a transformer top oil temperature prediction method is shown, including:

[0062] S101, obtain the first data of the target transformer, and perform a first preprocessing on the first historical data to obtain second data;

[0063] In the embodiment of this application, the first data of the target transformer is the historical monitoring data of the transformer oil temperature, and this historical monitoring data may contain various factors affecting the oil temperature, such as load rate, ambient temperature, cooling method, etc.

[0064] In an optional embodiment, the first data of the target transformer can be obtained by real-time monitoring through a temperature sensor installed on the transformer or extracted from the historical database. These first data may contain a large amount of noise and redundant information, so preprocessing is required to improve the data quality. This application does not limit the data acquisition method, and any method that can obtain the historical oil temperature data of the target transformer can be adopted.

[0065] It should be noted that the first preprocessing step aims to improve the quality and usability of the data through a series of technical means, so as to provide more accurate and meaningful input for subsequent oil temperature prediction.

[0066] In an optional embodiment, the first preprocessing can be performed by means of data cleaning, missing value filling, outlier handling, etc. to eliminate noise and redundant information in the data. In particular, considering the characteristics of the oil temperature data, methods such as moving average filtering and median filtering can be used to smooth the data and reduce the impact of random fluctuations on the prediction results. In addition, considering the possible non-linear characteristics and periodic changes of the oil temperature data, techniques such as phase space reconstruction and time series decomposition can also be used to further extract useful information in the data and provide more comprehensive feature inputs for the subsequent prediction model.

[0067] It should be noted that, in order to extract more accurate feature information, this application needs to decompose and reconstruct the first data, and use the reconstructed result as the input of the subsequent model to improve the prediction accuracy of the model.

[0068] In the embodiment of this application, the first preprocessing includes a first decomposition operation and a first reconstruction operation.

[0069] In the embodiment of this application, the first decomposition operation and the first reconstruction operation include:

[0070] The first decomposition operation is used to decompose the first data, and the first data is the oil temperature data of the target transformer;

[0071] The first reconstruction operation is used to reconstruct the first data decomposed by the first decomposition operation;

[0072] The first reconstruction operation extracts features from the first data decomposed by the first decomposition operation by calculating the distribution entropy to obtain the second data.

[0073] In an optional embodiment, the first decomposition operation can use signal decomposition algorithms in the prior art, such as empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), etc. These algorithms can decompose complex data sequences into a series of intrinsic mode function (IMF) components, and each IMF component represents different frequency components in the data, which helps to reduce data complexity and extract key features.

[0074] It should be noted that the reason for not using the above-mentioned existing technologies is that although the EMD algorithm can decompose data into multiple IMF components, there is a mode mixing problem, that is, there may be frequency overlap between different IMF components, resulting in inaccurate decomposition results. Although the EEMD algorithm alleviates the mode mixing problem to a certain extent by introducing white noise, the introduction of white noise may also contaminate the data and affect subsequent reconstruction and feature extraction. Therefore, the present invention adopts the CEEMD algorithm, that is, the complete ensemble empirical mode decomposition algorithm, which combines the advantages of EMD and EEMD and can avoid mode mixing and noise pollution problems while effectively decomposing data.

[0075] In the embodiment of the present application, the signal decomposition of the first data includes:

[0076] Adding positive and negative random Gaussian white noise to the Nth power to the first data;

[0077] Using the EMD algorithm to decompose the first data after adding white noise;

[0078] Taking the average value of each IMF component obtained by decomposition as the final decomposition result, and the final decomposition result is in the form of a data matrix.

[0079] It should be noted that the CEEMD algorithm overcomes the mode mixing problem by introducing white noise on the basis of the EMD algorithm, so as to obtain more accurate and stable IMF components. Specifically, positive and negative random Gaussian white noise to the Nth power is added to the first data, and then the EMD algorithm is used to decompose the first data after adding white noise. Finally, the average value of each IMF component obtained by decomposition is used as the final decomposition result. The final decomposition result obtained in this way is presented in the form of a data matrix and contains information on the first data at different frequency components.

[0080] In an alternative embodiment, some existing technologies can be used for the first reconstruction operation. For example, methods such as principal component analysis (PCA) and linear discriminant analysis (LDA) can be used for feature extraction to further reduce the data dimension and extract key features. However, these methods may not be able to fully capture the non-linear characteristics and potential laws in the oil temperature data. Therefore, the present invention adopts the DistEn data reconstruction method, which evaluates the complexity and uncertainty of the data by calculating the distribution entropy, so as to effectively extract the potential features and information in the oil temperature data.

[0081] In the embodiment of the present application, the feature extraction of the first data after the first decomposition operation by calculating the distribution entropy includes:

[0082] Calculating the maximum absolute difference between each vector in the final decomposition result and all other vectors;

[0083] Build a histogram based on the maximum absolute difference, and perform a segmentation operation on the histogram;

[0084] Calculate the number of occurrences and probability density of the maximum absolute error in each segmented region;

[0085] Calculate the distribution entropy according to the probability density, and complete the feature extraction of the first data after the first decomposition operation.

[0086] It should be noted that by calculating the distribution entropy to extract features from the first data after the first decomposition operation, the complexity and uncertainty of the data can be quantified, so as to extract more critical features for oil temperature prediction. These features can reflect the potential laws and changing trends in the oil temperature data, providing more accurate and meaningful inputs for the establishment of subsequent prediction models.

[0087] Exemplarily, the operation steps for the present application to perform first preprocessing on the first historical data to obtain second data are as follows:

[0088] Use the CEEMD decomposition method to decompose the obtained transformer oil temperature data (i.e., the first data), and use DistEn to reconstruct the decomposed component data to extract more accurate feature information (i.e., the second data). The specific process is as follows:

[0089] First, use the complementary ensemble empirical mode decomposition method (CEEMD) to decompose the collected historical top oil temperature data of the transformer. The specific decomposition process is as follows:

[0090] It should be noted that the complementary ensemble empirical mode decomposition (CEEMD) has the characteristics of small recovery error and significantly improving the mode mixing problem existing in the EMD decomposition algorithm. Compared with the empirical mode decomposition (EMD) and the ensemble empirical mode decomposition (EEMD), there is less white noise residue in the IMFs decomposed by CEEMD, and an ideal decomposition result can be provided with fewer accumulation times. Therefore, the present application uses CEEMD to decompose the transformer oil temperature data. As Figure 3 shown, first add positive and negative random Gaussian white noise to the power of N to the original oil temperature data (i.e., the first data) o(t). The specific process is as follows:

[0091]

[0092] In the formula, and respectively represent the i-th positive white noise sequence and negative white noise sequence, and respectively represent the oil temperature data after adding positive and negative white noise. t = 1,..., m represents the dimension of the oil temperature data, and each column of oil temperature data has m data points.

[0093] Furthermore, the EMD algorithm is used to decompose the signal after adding white noise, and the expression formula of the decomposed sequence is as follows:

[0094]

[0095] In the formula, R ij (t) represents the j-th IMF component obtained by decomposing the i-th signal, and r i represents the residue obtained by decomposing the i-th signal.

[0096] Furthermore, calculate the average value of each IMF component obtained by decomposition to obtain the final decomposition result of the CEEMD algorithm. The specific process is as follows:

[0097]

[0098] In the formula, N represents the number of white noises, and R j (t) represents the j-th (j = 1,..., n) component obtained by the final decomposition, and Se represents the final residue obtained by decomposition.

[0099] Therefore, the m×(n + 1)-dimensional transformer oil temperature component data IMF after decomposition can be obtained as follows:

[0100] IMF = [R1 R2 … R j Se] (j = 1,..., n) (4)

[0101] It should be noted that for more convenient representation, the Se sequence is represented by R j+1 here, and the changed component matrix is represented as follows:

[0102] IMF = [R1 R2 … R j R j+1 (j = 1,..., n) (5)

[0103] Furthermore, the distribution entropy is used to analyze the state space corresponding items of the time series of the decomposed data matrix to quantify the distribution characteristics of the distance between vectors. The specific process is as follows:

[0104] For each vector R a in the data matrix, calculate its maximum absolute difference d b from all other vectors R ab (a ≠ b). The specific representation is as follows:

[0105] d ab = max(|R a -R b |) (6)

[0106] Wherein, R a and R b respectively represent the a-th and b-th vectors in the oil temperature component data matrix; d ab vector R a and R b the maximum absolute difference between them;

[0107] Furthermore, all the calculated distances d ab are constructed into a histogram, and the specific construction process is as follows:

[0108] [q, x]=hist(Dab, B) (7)

[0109] Wherein, the parameter B represents the number of parts of the segmentation data of the histogram; D ab represents the one-dimensional vector composed of all the calculated distances d ab ; q represents the number of times d ab appears in each segmentation region of the histogram; x represents the segmentation point of the histogram.

[0110] Furthermore, for each segmentation region in the histogram, calculate the number of times q that d ab appears and calculate the probability density f, that is:

[0111]

[0112] Finally, calculate the distribution entropy DistEn according to the probability density f, and the formula is as follows:

[0113]

[0114] It should be noted that obtaining the first data of the target transformer and performing the first preprocessing on the first historical data can improve the accuracy and robustness of the prediction model. By performing CEEMD decomposition and DistEn reconstruction on the oil temperature data, the present invention can extract more accurate and meaningful feature information, which can fully reflect the potential laws and change trends in the oil temperature data. These high-quality feature inputs can provide more comprehensive information support for the subsequent prediction model, so that the model can more accurately capture the change characteristics of the oil temperature data, improve the prediction accuracy and stability. In addition, this preprocessing method also helps to enhance the generalization ability of the model, so that it can still maintain good prediction performance when facing different working conditions and change conditions.

[0115] S102, establish a first prediction model and a second correction model connected in series with the first prediction model, and denote the model connected in series as the third model;

[0116] In an alternative embodiment, the first prediction model can be established through machine learning algorithms, such as support vector machine (SVM), random forest, neural network, etc. These algorithms can learn the potential laws and trends of data from the extracted features, so as to realize the prediction of oil temperature.

[0117] In an alternative embodiment, the first prediction model can also be constructed through deep learning algorithms, such as long short-term memory network (LSTM), gated recurrent unit (GRU), etc. These algorithms have significant advantages in processing time series data, can capture the long-term dependencies in the oil temperature data, and improve the prediction accuracy.

[0118] In the embodiment of the present application, a transformer top oil temperature prediction model based on ORedRVFL (Outlier-Robust Ensemble Deep Random Vector Functional Link Network) is established as the first prediction model. This model combines the random vector functional link network (RVFL) and the ensemble learning method, and improves the model's ability to process abnormal data by introducing an outlier robust mechanism. The ORedRVFL model enhances the robustness and prediction performance of the model by integrating multiple RVFL sub-models and using outlier detection techniques to identify and process outliers in the training data.

[0119] It should be noted that the advantage of using the method of the present application to establish the first prediction model compared with other methods is that it can more effectively process outliers and noise in the oil temperature data, and improve the stability and reliability of the prediction results. In traditional oil temperature prediction models, outliers often have a negative impact on the training and prediction performance of the model, resulting in inaccurate or fluctuating prediction results. The ORedRVFL model can automatically detect and process outliers in the training data by introducing an outlier robust mechanism, thereby reducing the impact of outliers on the model performance and improving the prediction accuracy and stability. In addition, the ORedRVFL model also combines the random vector functional link network (RVFL) and the ensemble learning method, and has the advantages of fast training speed and strong generalization ability, and can better adapt to different working conditions and changing conditions, providing strong support for the accurate prediction of the transformer top oil temperature.

[0120] It should be noted that considering that there are often certain inherent errors in the model during prediction, resulting in some potential information in the prediction error not being fully utilized. Therefore, the present application proposes to further correct the preliminary prediction result, and thus proposes a second correction model.

[0121] In the embodiments of the present application, the second correction model is used to correct the prediction results of the first prediction model to improve the accuracy of prediction. The second correction model can be constructed based on different algorithms or models, such as regression models, support vector regression (SVR), or other machine learning algorithms. By establishing the second correction model, the prediction errors of the first prediction model can be analyzed and learned, so as to achieve fine adjustment of the prediction results. This cascaded model structure, that is, the third model, can make full use of the advantages of the first prediction model and the second correction model to improve the overall prediction performance.

[0122] In the embodiments of the present application, the second correction model cascaded with the first prediction model includes:

[0123] Obtain an error sequence, where the error sequence is the difference sequence between the predicted value of the first prediction model and the true oil temperature value of the target transformer;

[0124] Establish a second correction model based on the error sequence, where the input of the second correction model is the error sequence and the output is the error prediction value;

[0125] Cascade the second correction model to the output position of the first prediction model.

[0126] Exemplarily, as Figure 4 shown, Figure 4 shows the OredRVFL (Regularized Random Vector Functional Link) neural network model for oil temperature prediction proposed in this patent. The ORedRVFL model is obtained by introducing regularization and norm strategies into the edRVFL model to improve the edRVFL model. The model consists of an input layer, multiple hidden layers (NL hidden layers), a weight bias setting layer, and a result integrated output layer. The historical data of the transformer oil temperature first enters the model through the input layer; then, the input data undergoes nonlinear transformation and feature extraction in the hidden layer; secondly, the weights connecting the hidden layer nodes and the output layer are used to adjust the contribution of the hidden layer output to the final prediction result; finally, the outputs of all hidden layers are averaged by majority voting through an ensemble method to obtain the final prediction result.

[0127] The specific process of establishing the first prediction model and the second correction model cascaded with the first prediction model in the present application is as follows:

[0128] In the RVFL model, the data input in the input layer traverses through the nodes in the hidden layer, undergoes nonlinear mapping in the hidden layer, and the final result is output through the output function. The RVFL network can be described as follows:

[0129]

[0130] where g(·) is the activation function, and w j is the weight of the j-th hidden node. X = [x1, x2,..., x N is the input matrix of the transformed transformer oil temperature data components, b j is the threshold, and β j is the output layer weight.

[0131] It should be noted that, to reduce the time consumed in the iteration process, the parameters w j and b j of RVFL are randomly determined according to a given distribution function (such as Gaussian distribution), and the weight β j of the model depends on minimizing the system error.

[0132]

[0133] where B t contains (J + N) weights β j , P is the number of training samples, and d is the hidden node vector.

[0134] Furthermore, the deep random vector functional link network (dRVFL) is obtained by adding hidden layers on the basis of RVFL. The data input of one hidden layer is as shown in the equation (the bias is omitted here):

[0135]

[0136] where W represents the hidden layer weight, H represents the hidden layer output, and [H (i-1) ; X] represents the concatenation of the output of the previous hidden layer and the input features of the input layer.

[0137] It should be noted that edRVFL is obtained by adding an ensemble method to dRVFL. dRVFL is an ensemble learning model that combines the concepts of deep learning and RVFL to enhance the performance of the model. Matrix inversion is required to calculate the output weight β in dRVFL. The final output β in edRVFL is divided into several small βs for independent calculation, that is, each hidden layer calculates a small β, and the final output is obtained through majority voting or taking the average of the models.

[0138] Furthermore, the ORedRVFL model can be obtained by using regularization and norm to improve the robustness and generalization ability of the edRVFL model. Using the l0 norm to reflect sparsity rather than the l2 norm, and combining statistical theory, the problem of minimizing the structural risk and the empirical risk is transformed into the problem of minimizing the l2 norm of the output weight β, making the training error e sparse.

[0139]

[0140] Equation (12) constitutes a non-convex optimization problem. To simplify this problem, it can be transformed into a convex relaxation problem with sparse characteristics, making it easier to solve. Combining relevant theories, the l1 norm can be used to replace the l0 norm in the above formula, which not only ensures the sparsity of e but also guarantees that the overall minimization is convex.

[0141]

[0142] The above formula is a convex optimization constraint problem that can be solved by the augmented Lagrangian multiplier method (ALM). The augmented Lagrangian function can be expressed as follows:

[0143]

[0144] where λ is the augmented Lagrangian multiplier vector and μ is the penalty parameter.

[0145] Furthermore, in the first stage of the ALM iteration, its minimization can be achieved by continuously updating the two unknowns e and β. The specific iteration steps are as follows:

[0146]

[0147] Furthermore, e i+1 and β i+1 in the above equation can be solved by the following equations:

[0148] β i+1 =(H T H + 2 / CμI) -1 H T (y - e i + λ i / μ) (17)

[0149]

[0150] Furthermore, in the above equation, represents the multiplication between the corresponding elements. By substituting the equation into the following equation, the predicted result can be found:

[0151]

[0152] It should be noted that establishing a first prediction model and a second correction model connected in series with the first prediction model, and denoting the series-connected model as the third model can significantly improve the accuracy and stability of the prediction of the top oil temperature of the transformer. First, the first prediction model is constructed based on ORedRVFL. By integrating multiple RVFL sub-models and using outlier detection technology, the model's ability to process abnormal data is effectively improved, and the influence of outliers on the prediction results is reduced. Secondly, the second correction model further corrects the prediction error of the first prediction model. Through the learning and analysis of the error sequence, fine adjustment of the prediction results is achieved. This series-connected model structure, that is, the third model, not only makes full use of the advantages of the two models, but also enhances the overall prediction performance, enabling high accuracy and stability to be maintained even when facing complex and variable oil temperature data. In addition, this series-connected model also has good generalization ability and can adapt to different working conditions and changing conditions, providing a reliable guarantee for the accurate prediction of the top oil temperature of the transformer.

[0153] S103, preset a first improvement algorithm, and optimize the hyperparameters of the third model according to the first improvement algorithm, where the hyperparameters include a regularization factor and the number of hidden layer nodes;

[0154] In an optional embodiment, the algorithms for optimizing the hyperparameters of the third model may include but are not limited to grid search, random search, Bayesian optimization, etc. These algorithms can automatically search for the optimal combination of hyperparameters within the given parameter range, thereby further improving the prediction performance of the model.

[0155] It should be noted that grid search is an exhaustive search method that traverses the given parameter combinations and determines the optimal parameter settings through cross-validation. Random search is a heuristic search method that randomly samples within the parameter space and searches for the optimal solution by comparing the model performance under different parameter combinations.

[0156] It should be noted that Bayesian optimization is an optimization method based on a probability model. It uses historical search results to guide new parameter sampling and gradually approaches the optimal solution by iteratively updating the probability model.

[0157] In the embodiments of the present application, the improvement algorithms in the prior art are not used because although these algorithms can improve the prediction performance of the model to a certain extent, they often have problems such as large computational amount, slow convergence speed, or being easily trapped in local optimal solutions. To solve these problems, the present application proposes a new first improvement algorithm.

[0158] In the embodiments of the present application, the preset first improvement algorithm includes improving the differential evolution algorithm by introducing Tent initialization and the golden sine optimization strategy to obtain the first improvement algorithm.

[0159] In the embodiments of the present application, the hyperparameter optimization of the third model according to the first improved algorithm includes:

[0160] Establish an optimization objective function based on the third model;

[0161] Take the hyperparameters as the population type of the first improved algorithm;

[0162] Optimize the hyperparameters of the third model according to the first improved algorithm, and determine whether the termination condition is satisfied;

[0163] Output the hyperparameters that satisfy the termination condition, and update the third model.

[0164] Exemplarily, as Figure 5 shown, the specific steps for optimizing the hyperparameters of the third model according to the first improved algorithm in the present application are as follows:

[0165] It should be noted that the Differential Evolution (DE) algorithm is a population-based adaptive global optimization algorithm and belongs to one of the evolutionary algorithms. Its implementation steps are as follows:

[0166] (1) Initialization

[0167] First, initialize the population of the algorithm and assign a value to each dimension of each individual. After initialization, each individual is represented as follows:

[0168] x i,G (i = 1, 2,..., NP) (20)

[0169] In the formula, i represents the individual number, G represents the generation number, and NP is the population size.

[0170] In the differential evolution algorithm, assume the bounds of the parameter variables are At this time, the expression of x ji,0 is:

[0171]

[0172] (2) Mutation

[0173] After generating the initial population, perform the mutation operation. For each target x i,G (i = 1, 2,..., NP), the mutation vector generation method is as follows:

[0174]

[0175] In the formula, F is the mutation operator and F is a real constant factor between [1, 2], which is used to control the scaling of the deviation variable.

[0176] (3) Crossover

[0177] This process performs the overall crossover of the mutation vector and the target vector. To increase the diversity of the interference parameter vector, a crossover operation is introduced, and the test vector becomes:

[0178]

[0179] In the formula, rand(j) represents the j-th estimate of the random number generator, rnbr(i) represents a randomly selected sequence, and CR represents the crossover operator.

[0180] (4) Selection

[0181] The differential evolution algorithm compares the trial vector with the target vector xi,G in the current population according to the greedy criterion. In the next generation, if the target vector is good, the target vector is selected; if the trial vector is good, the trial vector is selected.

[0182] (5) Handling of boundary conditions

[0183] If, during the mutation process, a solution outside the feasible domain is compiled, the compiled vector is redefined within the feasible solution range as follows:

[0184]

[0185] Introducing Tent initialization and the Golden Sine Algorithm (Gold-SA) into DE to optimize and improve the initialization and position update phases of the DE algorithm, thereby obtaining the improved differential evolution algorithm TGDE; the specific process is as follows:

[0186] It should be noted that since the initialization process of the original algorithm is randomly given based on boundary conditions, it is very likely that the initialized population will be concentrated at a certain position, which may affect the overall convergence speed of the algorithm. To address the above issues, this application initializes the individuals in the algorithm using the Tent mapping with better randomness and uniformity, and the specific expression is as follows:

[0187]

[0188] x i,G = lb + (ub - lb)·r j (27)

[0189] In the formula, r j is a random number with a value in the range [0, 1], ψ is a chaotic parameter with a value in the range (0, 2], and lb and ub are the upper and lower limits of the solution range respectively.

[0190] It should be noted that the Gold-Sine Algorithm (Gold-SA) is derived from the sine function, where the number of search agents is evenly distributed in each dimension. This algorithm introduces the golden mean coefficient to reduce the search space and enhance the convergence of the algorithm. The core of the Gold-SA is its position update formula, which has strong local exploitation ability. The position update formula of Gold-SA is shown as follows:

[0191]

[0192] In the formula, r1 and r2 are random coefficients. r1 determines the distance that an individual travels in the next iteration, and r1 ∈ [0, 2π]. r2 determines the position update direction of the i-th individual in the next iteration, and r2 ∈ [0, π]. x1 and x2 are coefficients obtained by the golden section. The mathematical expressions of x1 and x2 are as follows:

[0193] x1 = ατ + β(1 - τ)

[0194] x2 = α(1 - τ) + βτ (29)

[0195] In the formula, α and β are search intervals, is the golden ratio coefficient.

[0196] It should be noted that considering that there are often certain inherent errors in the model during prediction, resulting in some potential information in the prediction error not being fully utilized. Therefore, this application proposes to further correct the preliminary prediction result. The specific steps of error correction are as follows:

[0197] Calculate the error sequence to obtain the error sequence, o i represents the i-th true value of the transformer oil temperature sequence, p i represents the i-th predicted value of the transformer oil temperature sequence, and the error sequence e i is expressed as follows.

[0198] e i = o i - p i (30)

[0199] Furthermore, train the ELM error model, reasonably divide the error sequence as the input of the error model for training and prediction, and obtain the error prediction value of the i-th sample.

[0200] Furthermore, add the i-th transformer oil temperature prediction value p i and the i-th error prediction value y i to obtain the i-th error-corrected transformer oil temperature prediction value, and finally obtain the final predicted sequence of the transformer oil temperature as shown in formula (31).

[0201]

[0202] Furthermore, the TGDE algorithm is used to optimize the parameters of the transformer top oil temperature prediction model based on ORedRVFL. The specific implementation process is as follows:

[0203] Taking the Root Mean Squared Error (RMSE) index between the predicted result of the transformer oil temperature and the real result as the objective function of the ORedRVFL oil temperature prediction model;

[0204]

[0205] In the formula, N is the total number of time steps, y t is the real transformer oil temperature value at time step t, is the transformer oil temperature value predicted by the model at time step t.

[0206] Furthermore, determine the regularization factor C and the number of hidden layer nodes Nn of the population type as the ORedRVFL model, which are specifically expressed as follows:

[0207]

[0208] Among them, X represents the population of the TGDE algorithm, and p represents the number of particles in one population.

[0209] It should be noted that a first improved algorithm is preset. Optimizing the hyperparameters of the third model according to the first improved algorithm can automatically search for and determine the optimal regularization factor C and the number of hidden layer nodes Nn, thus avoiding manual trial and error and the cumbersome parameter adjustment process. Through the optimization of the TGDE algorithm, the accuracy and generalization ability of the ORedRVFL oil temperature prediction model can be significantly improved. In addition, this optimization process can also improve the convergence speed of the model, reduce the training time, and make the model more efficient and reliable in practical applications. Therefore, using the TGDE algorithm to optimize the hyperparameters of the third model is one of the effective means to achieve accurate prediction of the transformer top oil temperature.

[0210] S104, using the second data as the input of the third model, and predicting the target transformer top oil temperature according to the output of the third model.

[0211] In an alternative embodiment, to make full use of historical data and real-time data and improve the accuracy of prediction, the second data, i.e., multi-dimensional data including historical oil temperature data, environmental parameters, load data, etc., can be used as the input of the third model. The third model, i.e., the cascaded prediction model after hyperparameter optimization, can output the predicted result of the top oil temperature of the target transformer based on these input data through its internal complex operation mechanism. This process not only considers the historical change trend of the transformer oil temperature but also incorporates the influence of real-time environmental parameters and load data, thus achieving a comprehensive and accurate prediction of the top oil temperature of the transformer. The prediction result can provide important references for the operation and dispatching of the power system, equipment maintenance, etc., and contribute to improving the security and stability of the power grid.

[0212] In summary, the present invention proposes a method for predicting the top oil temperature of a transformer, which uses the CEEMD algorithm to decompose the historical top oil temperature data of the transformer into multiple subsequences, reducing the data complexity and facilitating the prediction of subsequent models.

[0213] Using the DistEn data reconstruction method to reconstruct the top oil temperature data of the transformer can effectively extract the potential features and information in the top oil temperature data of the transformer, providing more accurate and meaningful input data for the subsequent prediction model, thereby improving the performance of the prediction model.

[0214] Aiming at the problems of slow convergence speed and easy getting stuck in local optimum in the optimization process of the differential evolution algorithm, the Tent initialization and Golden Sine optimization strategy (Gold-SA) are added to the initialization stage and population update process of the differential evolution algorithm to help the algorithm jump out of the local optimum, thus constructing the TGDE algorithm, enhancing the search ability of the original differential evolution algorithm and improving the search efficiency.

[0215] Aiming at the problems of high cost, limited response speed, dependence on a large amount of data and poor robustness in traditional methods for predicting the top oil temperature of a transformer, the present invention effectively simplifies the complexity of the model by using the ORedRVFL model, improving the robustness of the model while enhancing the efficiency; and uses TGDE to optimize the parameters in the ORedRVFL model, constructing the CEEMD-DistEn-TGDE-ORedRVFL-EC model, improving the performance and accuracy of the top oil temperature prediction model of the transformer. This method can accurately predict the top oil temperature of the transformer under small sample conditions, providing a strong guarantee for the safe operation of the power system.

[0216] Embodiment 2 Figure 6This is the fitted line graph of the CEEMD-DistEn-TGDE-ORedRVFL-EC transformer oil temperature prediction model proposed in this application for the predicted results of the transformer oil temperature in summer and winter after adding random noise interference. Among them Figure 6 (a) is the fitted graph of the predicted results of the summer oil temperature, Figure 6 (b) is the fitted graph of the predicted results of the winter oil temperature.

[0217] It can be clearly seen from the fitted graph that Figure 6 (a) shows that the oil temperature data in summer exhibits several obvious peaks and has large overall fluctuations. The CEEMD-DistEn-TGDE-ORedRVFL-EC transformer oil temperature prediction model proposed in this patent shows a high degree of coincidence between its prediction curve and the true value curve in both summer and winter, and can capture the main fluctuation characteristics of the true value well. This proves that the prediction model proposed in this patent has high accuracy and robustness. In addition, the prediction effect of the prediction model of this application on the transformer oil temperature in winter is slightly better than that in summer as a whole. This is because the large volatility of electricity consumption in summer leads to complex and changeable transformer oil temperature and low regularity, which further confirms the authenticity of the oil temperature data in this article.

[0218] Example 3. In this example, a transformer top oil temperature prediction system is also provided, including:

[0219] A preprocessing module for obtaining the first data of the target transformer and performing a first preprocessing on the first historical data to obtain the second data;

[0220] The first preprocessing includes a first decomposition operation and a first reconstruction operation;

[0221] A model establishment module for establishing a first prediction model and a second correction model connected in series with the first prediction model, and denoting the series-connected model as the third model;

[0222] A model optimization module for presetting a first improvement algorithm and performing hyperparameter optimization on the third model according to the first improvement algorithm, where the hyperparameters include a regularization factor and the number of hidden layer nodes;

[0223] A model prediction module for using the second data as the input of the third model and predicting the top oil temperature of the target transformer according to the output of the third model.

[0224] The above-mentioned unit modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form, so as to facilitate the processor to call and execute the operations corresponding to the above-mentioned modules.

[0225] This embodiment also provides a computer device, which can be a terminal, and its internal structure diagram can be as shown in Figure 7 . The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, 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 implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. When the computer program is executed by the processor, it implements a method for predicting the top oil temperature of a transformer. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0226] This embodiment also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0227] Obtain the first data of the target transformer, and perform a first preprocessing on the first historical data to obtain second data;

[0228] The first preprocessing includes a first decomposition operation and a first reconstruction operation;

[0229] Establish a first prediction model and a second correction model connected in series with the first prediction model, and denote the series-connected model as the third model;

[0230] Preset a first improvement algorithm, and optimize the hyperparameters of the third model according to the first improvement algorithm. The hyperparameters include a regularization factor and the number of hidden layer nodes;

[0231] Use the second data as the input of the third model, and predict the top oil temperature of the target transformer according to the output of the third model.

[0232] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

[0233] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code. The solutions in the embodiments of the present application can be implemented using various computer languages.

[0234] The present application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0235] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufacture including instruction means that implement the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0236] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0237] Although the preferred embodiments of the present application have been described, those skilled in the art can make additional changes and modifications once they learn the basic creative concepts. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0238] Obviously, those skilled in the art can make various changes and modifications to this application without departing from the spirit and scope of this application. Thus, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalent technologies, this application is also intended to include these changes and modifications.

Claims

1. A method for predicting the top oil temperature of a transformer, characterized in that, Including: Obtain the first data of the target transformer, and perform a first preprocessing on the first historical data to obtain second data; The first preprocessing includes a first decomposition operation and a first reconstruction operation; Establish a first prediction model and a second correction model connected in series with the first prediction model, and denote the series-connected model as the third model; Preset a first improvement algorithm, and optimize the hyperparameters of the third model according to the first improvement algorithm, where the hyperparameters include a regularization factor and the number of hidden layer nodes; Use the second data as the input of the third model, and predict the top oil temperature of the target transformer according to the output of the third model.

2. The transformer top oil temperature prediction method according to claim 1, characterized in that The first decomposition operation and the first reconstruction operation include: The first decomposition operation is used to decompose the first data, and the first data is the oil temperature data of the target transformer; The first reconstruction operation is used to reconstruct the first data decomposed by the first decomposition operation; The first reconstruction operation extracts features from the first data decomposed by the first decomposition operation by calculating the distribution entropy to obtain second data.

3. The transformer top oil temperature prediction method according to claim 1 or 2, characterized in that The second correction model connected in series with the first prediction model includes: Obtain an error sequence, where the error sequence is the difference sequence between the predicted value of the first prediction model and the true oil temperature value of the target transformer; Establish a second correction model based on the error sequence, where the input of the second correction model is the error sequence and the output is the error prediction value; Connect the second correction model in series to the output position of the first prediction model.

4. The transformer top oil temperature prediction method according to claim 3, characterized in that The preset first improvement algorithm includes improving the differential evolution algorithm by introducing Tent initialization and the golden sine optimization strategy to obtain the first improvement algorithm.

5. The transformer top oil temperature prediction method according to claim 1 or 4, characterized in that The optimizing the hyperparameters of the third model according to the first improvement algorithm includes: Establish an optimization objective function based on the third model; Use the hyperparameters as the population type of the first improvement algorithm; Optimize the hyperparameters of the third model according to the first improvement algorithm, and determine whether the termination condition is satisfied; Output the hyperparameters that satisfy the termination condition and update the third model.

6. The transformer top oil temperature prediction method according to claim 2, characterized in that, The decomposing the first data into signals includes: Add positive and negative random Gaussian white noise to the Nth power of the first data; Use the EMD algorithm to decompose the first data after adding white noise; Take the average value of each IMF component obtained by decomposition as the final decomposition result, and the final decomposition result is in the form of a data matrix.

7. The transformer top oil temperature prediction method according to claim 2, characterized in that The extracting features from the first data decomposed by the first decomposition operation by calculating the distribution entropy includes: Calculate the maximum absolute difference between each vector in the final decomposition result and all other vectors; Establish a histogram based on the maximum absolute difference and perform a segmentation operation on the histogram; Calculate the number of occurrences and the probability density of the maximum absolute error in each segmentation region; Calculate the distribution entropy according to the probability density to complete the feature extraction of the first data decomposed by the first decomposition operation.

8. A transformer top oil temperature prediction system, applied to the method according to any one of claims 1 to 7, characterized in that, Including: A preprocessing module, configured to obtain the first data of the target transformer and perform a first preprocessing on the first historical data to obtain second data; The first preprocessing includes a first decomposition operation and a first reconstruction operation; A model establishment module, configured to establish a first prediction model and a second correction model connected in series with the first prediction model, and denote the series-connected model as a third model; A model optimization module, configured to preset a first improvement algorithm and perform hyperparameter optimization on the third model according to the first improvement algorithm, where the hyperparameters include a regularization factor and the number of hidden layer nodes; A model prediction module, configured to use the second data as the input of the third model and perform target transformer top oil temperature prediction according to the output of the third model.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the transformer top oil temperature prediction method according to any one of claims 1 to 7 are implemented.

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 top oil temperature prediction method according to any one of claims 1 to 7 are implemented.

Images