Predictive control method and device for heavy overload top oil temperature of transformer and storage medium
Through the parallel CNN-BiGRU model and Meyer-KAN prediction model, the spatial static characteristics and timing characteristics of the transformer oil temperature are extracted, and the problem of low oil temperature prediction accuracy in the existing technology is solved, achieving higher prediction accuracy and adaptability.
Patent Information
- Application Number
- CN202411484237.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-23
- Publication Date
- 2025-02-07
AI Technical Summary
The prior art is difficult to effectively extract the spatial and timing characteristics of the transformer oil temperature sequence under heavy overload conditions, resulting in low oil temperature prediction accuracy.
The parallel CNN-BiGRU model and Meyer-KAN prediction model are adopted to extract spatial static features and timing features through convolutional neural networks and gated cyclic units, and combined with Meyer wavelet activation function, the model's learning ability and prediction accuracy are enhanced.
It improves the accuracy and generalization ability of transformer oil temperature prediction, can effectively integrate multivariate data, and adapt to oil temperature changes under heavy overload conditions.
Smart Images
Figure CN119395991A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of power transmission and transformation, and relates to a method, a device and a storage medium for predicting and controlling the oil temperature of the top layer of a transformer under heavy overload. Background Art
[0002] As one of the key equipment for power transmission, the operating condition of power transformers has a great impact on the safety of the power grid. With the continuous improvement of my country's urbanization level, the demand for urban electricity has increased year by year, while the number of transformers is relatively limited, resulting in a continuous increase in the load rate of transformers. Especially in hot summer weather and central heating in winter, due to the sharp increase in the number of loads, cases of transformer overload and heavy load are common. Transformer overload leads to a decline in transformer performance and increased internal heat. Once the temperature exceeds the rated temperature range of the transformer, it may cause damage to the insulation material or cooling system and shorten the life of the transformer. Therefore, it is very important to monitor the condition of the transformer under heavy overload conditions.
[0003] The insulating oil temperature at the top of the transformer tank is an important indicator for measuring the insulation state of the transformer. If the top oil temperature is too hot, it will affect the insulation strength of the transformer. Therefore, the prediction of the top oil temperature of the transformer can provide an important basis for transformer condition monitoring. At present, the prediction of the top oil temperature of the transformer is mainly based on numerical models, semi-physical models and data-driven models. Numerical models are usually based on physical laws and thermodynamic principles to construct mathematical equations for heat transfer inside the transformer to predict the top oil temperature. This method requires transformer structural parameters and has a large amount of calculation, so it is not suitable for transformer top oil temperature monitoring. The semi-physical model based on heat transfer has the disadvantages of large errors and difficulty in obtaining heat transfer parameters, and is not widely used. Data-driven machine learning methods provide a new solution for intelligent prediction of transformer oil temperature, such as support vector machines and artificial neural networks. These methods show good prediction accuracy and generalization ability. However, machine learning methods do not fully consider the impact of different operating parameters and have not yet mined the correlation between data.
[0004] In recent years, with the continuous development of deep learning methods, algorithms such as recurrent neural networks (RNN) and convolutional neural networks (CNN) have emerged. With their outstanding nonlinear fitting capabilities, they have achieved great results in the field of oil temperature prediction. However, there are still some problems. First, the current transformer oil temperature prediction model mostly uses a single variable, without comprehensively considering the influence of load, weather conditions and other oil temperature related factors. Second, the oil temperature features extracted by the current prediction model are single and cannot fully characterize the spatial and temporal characteristics of the oil temperature sequence. Therefore, a single deep learning algorithm is difficult to effectively predict the transformer oil temperature. Summary of the invention
[0005] The technical solution of the present invention is used to solve the problem that the transformer oil temperature prediction model using a single variable in the prior art is difficult to effectively extract the spatial characteristics and temporal characteristics that characterize the oil temperature sequence under heavy overload conditions.
[0006] The present invention solves the above technical problems through the following technical solutions:
[0007] The present invention provides a transformer heavy overload top layer oil temperature prediction and control method, comprising the following steps:
[0008] Collect top layer oil temperature data, perform outlier detection and normalization on the original data, and select training sets and test sets in proportion from the normalized data;
[0009] A parallel CNN-BiGRU model and a Meyer-KAN prediction model were established, and the training set was input into the parallel CNN-BiGRU model and the Meyer-KAN prediction model for training. The data were respectively extracted from the spatial static features and the temporal features through the convolutional neural network and the gated recurrent unit. The extracted spatial static features and temporal features were respectively given different weights for weighting, and the weights were adjusted according to the error between the true value and the predicted value.
[0010] Input the data of the validation set into the trained model to observe whether the prediction accuracy of the validation set meets the requirements. If the prediction accuracy meets the requirements, the model is used for the test set; if the prediction accuracy does not meet the set value, adjust the parameters and train again; use the trained model to predict the test samples and calculate the error evaluation index at the same time;
[0011] Set the warning value and limit value of the transformer oil temperature under heavy load conditions. If the predicted value obtained by the model is greater than the warning value of the transformer oil temperature, adjust the operating frequency of the cooling equipment to reduce the rate of oil temperature rise; if the predicted value of the model is greater than the oil temperature limit value, reduce the load of the transformer or shut down to ensure equipment safety.
[0012] Preferably, the outlier detection adopts the 3σ criterion for outlier detection, and the formula for outlier detection using the 3σ criterion is as follows:
[0013] P(|x-μ|>3σ)≤0.003 (1)
[0015] Where x represents a specific observation value in the original data, μ and σ are the mean and standard deviation of the original data, and P(·) represents the probability of an event occurring.
[0016] Furthermore, the formula for the normalization process is as follows:
[0017]
[0018] In the formula, X represents the data before normalization, represents the normalized data, X max is the maximum value in historical data, X min This is the minimum value in historical data.
[0019] Furthermore, the method for establishing the parallel CNN-BiGRU model is as follows:
[0020] The one-dimensional convolution calculation formula is as follows:
[0021]
[0022] In the formula, w k is the convolution kernel, x t-k+1 is a time series, y t is the output feature after convolution;
[0023] The operation steps of the gated recurrent unit are as follows:
[0024] z t =σ(W z x t +U z h t-1 ) (4)
[0026] r t =σ(W r x t +U r h t-1 ) (5)
[0028]
[0029] Among them, z t and r t They represent the states of the update gate and the reset gate at time t, respectively. t-1 represents the input at the previous moment, h t Represents the output, x t Indicates the input data at the current moment, W z , W r , W h , U z , U r , U h All represent weight matrices, σ represents the sigmoid function, represents the Hadamard product, represents the candidate hidden state, h t Indicates the final output state;
[0030] A bidirectional gated loop unit is used to obtain information between data from both the front and back directions to predict the current oil temperature state. The specific formula is as follows:
[0031]
[0032] In the formula, h t is the output feature of BiGRU, GRU(·) represents the gated recurrent unit, and is the output of the forward and backward hidden layers, α t and β t are the weights of the forward and backward hidden layers, b t is the bias, and Represent the outputs of the forward and backward hidden layers at time t-1 respectively.
[0033] Furthermore, the formula for weighting the extracted spatial static features and temporal features by assigning different weights is as follows:
[0034] F = W1 × maxpooling (y t )+W2×h t (9)
[0035] Where F is the weighted feature extracted by the CNN-BiGRU model, and W1 and W2 are weights.
[0036] Furthermore, the method for establishing the Meyer-KAN prediction model is as follows:
[0037] Any multivariable continuous function is expressed as a combination of a series of single variable functions, and the formula is as follows:
[0038]
[0039] Among them, φ q,p : and Φ q φ q,p is to map each input variable x p A single variable function of ;
[0040] For an L-layer KAN network, the output of the KAN network is defined as f(x)≡KAN(x), then,
[0041]
[0042] Among them, n L-1 is the number of nodes in the L-1 layer of the KAN network, Represents the i-th layer of L-1 L-1 nodes and L layers iL The activation function between nodes, Represents input features;
[0043] Meyer wavelet is used as the activation function of the KAN network. The scale factor and displacement factor in the wavelet transform are optimized by gradient descent to minimize the loss function. The expression of Meyer wavelet is defined as follows:
[0044]
[0045] Where v(x) = x 4 (35-84x+70x 2 -20x 3 ),x∈(0,1),ψ(ω) is the frequency domain expression of Meyer wavelet, ω is the frequency variable, which is used to describe the frequency component of the signal.
[0046] Preferably, the error evaluation indicators include: mean absolute percentage error, root mean square error, and determination coefficient.
[0047] Furthermore, the calculation formula of the mean absolute percentage error is as follows:
[0048]
[0049] The calculation formula of the root mean square error is as follows:
[0050]
[0051] The calculation formula of the determination coefficient is as follows:
[0052]
[0053] In the formula, n is the number of samples, y is i is the actual value, is the predicted value, is the mean of the actual values.
[0054] The present invention also provides a device, including a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the above-mentioned transformer heavy overload top layer oil temperature prediction and control method, and the processor is configured to execute the program stored in the memory.
[0055] The present invention also provides a storage medium having a computer program stored thereon. When the computer program is run by a processor, the steps of the above-mentioned transformer heavy overload top layer oil temperature prediction and control method are executed.
[0056] The advantages of the present invention are:
[0057] The present invention utilizes CNN network to extract spatial static features of oil temperature prediction data, utilizes BiGRU algorithm to perform deep mining on oil temperature prediction data in both directions, and combines Meyer-KAN prediction model to strengthen the mutual connection between different time intervals of the same feature and different features in the same time interval, thereby enhancing the learning ability of the model and improving the oil temperature prediction accuracy. It can effectively integrate multivariate data to predict the oil temperature of transformer under heavy overload conditions, and has good accuracy and certain generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] Figure 1 It is a flow chart of a transformer heavy overload top layer oil temperature prediction and control method according to a first embodiment of the present invention;
[0059] Figure 2 It is the structural diagram of the CNN network;
[0060] Figure 3 It is the network structure diagram of GRU;
[0061] Figure 4 1 is a network structure diagram of a BiGRU model of a transformer heavy overload top layer oil temperature prediction and control method according to a first embodiment of the present invention;
[0062] Figure 5 It is a network structure diagram of a parallel CNN-BiGRU model of a transformer heavy overload top layer oil temperature prediction and control method according to the first embodiment of the present invention;
[0063] Figure 6 It is a network structure diagram of the Meyer-KAN prediction model of the transformer heavy overload top layer oil temperature prediction control method according to the first embodiment of the present invention;
[0064] Figure 7 It is a prediction curve of the parallel CNN-BiGRU model and the Meyer-KAN prediction model of the transformer heavy overload top layer oil temperature prediction control method of the first embodiment of the present invention. DETAILED DESCRIPTION
[0065] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0066] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments:
[0067] Embodiment 1
[0068] like Figure 1 As shown, a method for predicting and controlling the top layer oil temperature under a transformer heavy overload condition according to an embodiment of the present invention comprises the following steps:
[0069] 1. Data collection, preprocessing and data set division
[0070] The top layer oil temperature data was collected, and the original data were subjected to outlier detection and normalization to ensure that the data were on the same scale.
[0071] 1) Outlier detection: There are outliers in the oil temperature prediction data set, which will affect the prediction accuracy. Since the oil temperature prediction data is approximately normally distributed, the 3σ criterion is used for outlier detection. The formula for outlier detection using the 3σ criterion is as follows:
[0072] P(|x-μ|>3σ)≤0.003 (1)
[0074] Where x represents a specific observation value in the original data, μ and σ are the mean and standard deviation of the original data, and P(·) represents the probability of an event occurring.
[0075] In a set of data, if a measurement value differs from the mean by more than 3 standard deviations, the value is considered an outlier.
[0076] 2) Normalization: In order to eliminate the impact of different dimensions between input variables, the input data is normalized. The normalization formula is as follows:
[0077]
[0078] In the formula, X represents the data before normalization, represents the normalized data, X max is the maximum value in historical data, X min This is the minimum value in historical data.
[0079] 3) Randomly select 70% of the normalized data for training, and the remaining 30% is used as a test set to verify the accuracy and generalization ability of the model.
[0080] 2. Building a parallel CNN-BiGRU model
[0081] The training set is input into the parallel CNN-BiGRU model, and the spatial static features and temporal features of the data are extracted through the convolutional neural network and the gated recurrent unit respectively. The extracted spatial static features and temporal features are weighted by assigning different weights respectively.
[0082] 1) Parallel CNN-BiGRU model
[0083] The convolutional neural network (CNN) model uses local connections and weight sharing to process the original data into high-dimensional maps and effectively extract data features.
[0084] like Figure 2 As shown in the figure, the structure of the CNN network includes convolutional layers, pooling layers, and fully connected layers. The convolutional neural network first extracts the features of the input data through the convolutional layer, then refines these features through the pooling layer, and finally outputs them through the fully connected layer.
[0085] The one-dimensional convolution calculation formula is as follows:
[0086]
[0087] In the formula, w k is the convolution kernel, x t-k+1 is a time series, y t is the output feature after convolution.
[0088] The structure of the gated recurrent unit (GRU) is as follows Figure 3 As shown in the figure, the gated recurrent unit is evolved from LSTM and mainly consists of an update gate and a reset gate. The reset gate is used to obtain long-term relationships, while the update gate can obtain short-term relationships. It has good time-related sequence capabilities.
[0089] The operation steps of the gated recurrent unit are as follows:
[0090] z t =σ(W z x t +U z h t-1 ) (4)
[0092] r t =σ(W r x t +U r h t-1 ) (5)
[0094]
[0095] Among them, z t and r t They represent the states of the update gate and the reset gate at time t, respectively. t-1 represents the input at the previous moment, h t Represents the output, x t Indicates the input data at the current moment, W z , W r , W h , U z , Ur , U h Both represent weight matrices, σ represents the sigmoid function, represents the Hadamard product, represents the candidate hidden state, h t Indicates the final output status.
[0096] In order to fully consider the information law of the changes between the data, a bidirectional GRU (BiGRU) structure is adopted. BiGRU consists of two unidirectional GRUs in opposite directions. The calculation formula is as follows: Figure 4 As shown in the figure, the output of the model is determined by the weighted sum of the results of a forward hidden layer and a backward hidden layer. Applying it to prediction can obtain information between data from both the forward and backward directions, predict the current oil temperature state, and improve the performance of the model.
[0097]
[0098] In the formula, h t is the output feature of BiGRU, GRU(·) represents the gated recurrent unit, and is the output of the forward and backward hidden layers, α t and β t are the weights of the forward and backward hidden layers, b t is the bias, and Represent the outputs of the forward and backward hidden layers at time t-1 respectively.
[0099] The network structure of the parallel CNN-BiGRU spatiotemporal feature extraction is as follows Figure 5 As shown. The original data is extracted through CNN and BiGRU to extract spatial static features and time series features, and the extracted two parts of features are weighted by assigning different weights. The formula is as follows:
[0100] F = W1 × maxpooling (y t )+W2×h t (9)
[0101] Where F is the weighted feature extracted by the CNN-BiGRU model, and W1 and W2 are weights.
[0102] 3. Establish Meyer-KAN prediction model
[0103] KAN is a new neural network architecture inspired by the Kolmogorov-Arnold representation theorem, which states that any multivariable continuous function can be represented as a combination of a series of single-variable functions. The formula is as follows:
[0104]
[0105] Among them, φ q,p : and Φ:
[0106] q φ q,p is to map each input variable x p A single variable function of .
[0107] KAN network is different from the traditional neural network that uses fixed activation function. It uses learnable activation function at the edge of the network. The traditional weight parameters are replaced by single variable function parameters at the edge of the network. For an L-layer KAN network, the output of KAN is defined as f(x)≡KAN(x), then,
[0108]
[0109] Among them, n L-1 is the number of nodes in the L-1 layer of the KAN network, Represents the i-th layer of L-1 L-1 nodes and L layers i L The activation function between nodes, Represents the input features.
[0110] The activation function of the traditional KAN network usually uses the B-spline function, but the feature extraction capability of the spline function is limited, and the computational complexity increases significantly with the increase of dimension, making the KAN network not suitable for oil temperature prediction data. Wavelets are good at multi-resolution, capturing high-frequency details and low-frequency trends of the data. In addition, wavelets provide sparse representation, which can lead to more efficient neural network architectures and faster training times. It is very suitable for processing non-stationary signals and local features. The Meyer wavelet is infinitely differentiable and tightly supported in the frequency domain, and no aliasing will occur in the frequency domain. In addition, the Meyer wavelet is very smooth and has good wave characteristics. The expression of the Meyer wavelet is defined as follows:
[0111]
[0112] Where v(x) = x 4 (35-84x+70x 2 -20x 3 ),x∈(0,1),ψ(ω) is the frequency domain expression of Meyer wavelet, ω is the frequency variable, which is used to describe the frequency component of the signal.
[0113] In the Meyer-KAN prediction model, Meyer wavelet is used as the activation function of the network. During the training process, the scale factor and displacement factor in the wavelet transform are optimized by gradient descent to minimize the loss function. Each iteration updates the wavelet parameters to reduce the prediction error. The network structure of the Meyer-KAN prediction model is as follows: Figure 6 As shown. In this embodiment, the KAN model has a total of 3 layers, the number of nodes in the middle hidden layer is 32, the number of iterations is 100, the batch size is 32, and the initial learning rate is 0.001. The data of the validation set is input into the trained model to observe whether the prediction accuracy of the validation set meets the requirements. If the prediction accuracy meets the requirements, the model is used for the test set; if the prediction accuracy does not meet the set value, the parameters are adjusted and trained again; the trained model is used to predict the test samples, and different error evaluation indicators are calculated at the same time; the error evaluation indicators include mean absolute percentage error (MAPE), root mean square error (RMSE), determination coefficient (R 2 ).
[0114] The calculation formula of mean absolute percentage error (MAPE) is as follows:
[0115]
[0116] The root mean square error (RMSE) is calculated as follows:
[0117]
[0118] Coefficient of determination (R 2 ) is calculated as follows:
[0119]
[0120] In the formula, n is the number of samples, y is i is the actual value, is the predicted value, is the mean of the actual values.
[0121] 4. Control strategy
[0122] According to the transformer design specifications, the temperature resistance of the insulation material and the use environment, the warning value and limit value of the transformer oil temperature under heavy load conditions are set respectively. If the predicted value obtained by the Meyer-KAN prediction model is greater than the warning value of the transformer oil temperature, the operating frequency of the cooling equipment is adjusted to reduce the oil temperature rise rate. If the model's predicted value is greater than the oil temperature limit value, the transformer load is gradually reduced or shut down to ensure equipment safety.
[0123] 5. Test verification
[0124] like Figure 7As shown in the figure, the transformer top layer oil temperature prediction curve obtained by experiment, it can be seen from the figure that the method of the present invention can better track the changing trend of the oil temperature at most time points, indicating that the method of the present invention has strong prediction accuracy and robustness; in the area where the oil temperature fluctuates significantly, the predicted value can effectively respond and keep consistent with the actual value, which reflects the adaptability of the method of the present invention in heavy load and complex environment.
[0125] In order to verify the superiority of the method of the present invention, it is compared with the representative methods such as TCN, CNN-LSTM and CNN-GRU. The MAPE, RMSE and R 2 The index results are shown in Table 1. As can be seen from the table, the method of the present invention shows obvious advantages in all indicators. In terms of MAPE, it is reduced by 37.5% compared with the TCN model, and by 17.5% and 16.6% compared with CNN-LSTM and CNN-GRU respectively. In terms of RMSE index, the error of the parallel CNN-BiGRU and Meyer-KAN networks is reduced to 0.2846, which is 49% less than TCN, and 35% and 29% less than CNN-LSTM and CNN-GRU respectively. In addition, in R 2 The method of the present invention achieves 0.9910, which is a significant improvement over other models. These results show that the method of the present invention can not only more accurately capture the complex spatiotemporal characteristics in the data, but also has stronger generalization ability and can better adapt to the prediction task of heavy overload oil temperature changes.
[0126] Table 1 MAPE, RMSE, R of different methods 2 Indicator results
[0127]
[0128] Embodiment 2
[0129] An electronic device includes a memory and a processor, wherein the memory is used to store a program that supports the processor to execute the transformer heavy overload top layer oil temperature prediction and control method in embodiment 1, and the processor is configured to execute the program stored in the memory.
[0130] Embodiment 3
[0131] A storage medium stores a computer program, which, when executed by a processor, executes the steps of the transformer heavy overload top layer oil temperature prediction and control method in embodiment 1.
[0132] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A transformer heavy overload top oil temperature prediction and control method, characterized in that: The following steps are involved: Collect top layer oil temperature data, perform outlier detection and normalization on the original data, and select training sets and test sets in proportion from the normalized data; A parallel CNN-BiGRU model and a Meyer-KAN prediction model were established, and the training set was input into the parallel CNN-BiGRU model and the Meyer-KAN prediction model for training. The data were respectively extracted from the spatial static features and the temporal features through the convolutional neural network and the gated recurrent unit. The extracted spatial static features and temporal features were respectively given different weights for weighting, and the weights were adjusted according to the error between the true value and the predicted value. Input the data of the validation set into the trained model to observe whether the prediction accuracy of the validation set meets the requirements. If the prediction accuracy meets the requirements, the model is used for the test set; if the prediction accuracy does not meet the set value, adjust the parameters and train again; use the trained model to predict the test samples and calculate the error evaluation index at the same time; Set the warning value and limit value of the transformer oil temperature under heavy load conditions. If the predicted value obtained by the model is greater than the warning value of the transformer oil temperature, adjust the operating frequency of the cooling equipment to reduce the rate of oil temperature rise; if the predicted value of the model is greater than the oil temperature limit value, reduce the load of the transformer or shut down to ensure equipment safety.
2. The transformer heavy overload top oil temperature prediction and control method according to claim 1 is characterized in that: The outlier detection adopts the 3σ criterion for outlier detection. The formula for outlier detection using the 3σ criterion is as follows: P(|x-μ|>3σ)≤0.003 (1) Where x represents a specific observation value in the original data, μ and σ are the mean and standard deviation of the original data, and P(·) represents the probability of an event occurring.
3. The transformer heavy overload top oil temperature prediction and control method according to claim 1 is characterized in that: The formula for the normalization process is as follows: In the formula, X represents the data before normalization, represents the normalized data, X max is the maximum value in historical data, X min This is the minimum value in historical data.
4. The transformer heavy overload top oil temperature prediction and control method according to claim 1 is characterized in that: The method for establishing the parallel CNN-BiGRU model is as follows: The one-dimensional convolution calculation formula is as follows: In the formula, w k is the convolution kernel, x t-k+1 is a time series, y t is the output feature after convolution; The operation steps of the gated recurrent unit are as follows: z t =σ(W z x t +U z h t-1 ) (4)r t =σ(W r x t +U r h t-1 ) (5) Among them, z t and r t They represent the states of the update gate and the reset gate at time t, respectively. t-1 represents the input at the previous moment, h t Represents the output, x t Indicates the input data at the current moment, W z , W r , W h , U z , U r , U h All represent weight matrices, σ represents the sigmoid function, represents the Hadamard product, represents the candidate hidden state, h t Indicates the final output state; A bidirectional gated loop unit is used to obtain information between data from both the front and back directions to predict the current oil temperature state. The specific formula is as follows: In the formula, h t is the output feature of BiGRU, GRU(·) represents the gated recurrent unit, and is the output of the forward and backward hidden layers, α t and β t are the weights of the forward and backward hidden layers, b t is the bias, and Represent the outputs of the forward and backward hidden layers at time t-1 respectively.
5. The transformer heavy overload top oil temperature prediction and control method according to claim 4 is characterized in that: The formula for weighting the extracted spatial static features and temporal features by assigning different weights is as follows: F=W1×maxpooling(y t )+W2×h t (9) Where F is the weighted feature extracted by the CNN-BiGRU model, and W1 and W2 are weights.
6. The transformer heavy overload top oil temperature prediction and control method according to claim 1 is characterized in that: The method for establishing the Meyer-KAN prediction model is as follows: Any multivariable continuous function is expressed as a combination of a series of single variable functions, and the formula is as follows: in, and φ q,p is to map each input variable x p A single variable function of ; For an L-layer KAN network, the output of the KAN network is defined as f(x)≡KAN(x), then, Among them, n L-1 is the number of nodes in the L-1 layer of the KAN network, Represents the i-th layer of L-1 L-1 nodes and L layers i L The activation function between nodes, Represents input features; Meyer wavelet is used as the activation function of the KAN network. The scale factor and displacement factor in the wavelet transform are optimized by gradient descent to minimize the loss function. The expression of Meyer wavelet is defined as follows: Where v(x) = x 4 (35-84x+70x 2 -20x 3 ),x∈(0,1),ψ(ω) is the frequency domain expression of Meyer wavelet, ω is the frequency variable, which is used to describe the frequency component of the signal.
7. The transformer heavy overload top oil temperature prediction and control method according to claim 1 is characterized in that: The error evaluation indicators include: mean absolute percentage error, root mean square error, and determination coefficient.
8. The transformer heavy overload top oil temperature prediction and control method according to claim 7 is characterized in that: The calculation formula for the mean absolute percentage error is as follows: The calculation formula of the root mean square error is as follows: The calculation formula of the determination coefficient is as follows: In the formula, n is the number of samples, y is i is the actual value, is the predicted value, is the mean of the actual values.
9. A device comprising a memory and a processor, characterized in that: The memory is used to store a program that supports the processor to execute the transformer heavy overload top layer oil temperature prediction and control method as described in any one of claims 1 to 8, and the processor is configured to execute the program stored in the memory.
10. A 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 heavy overload top layer oil temperature prediction and control method according to any one of claims 1 to 8 are executed.