Training method of oil temperature prediction model and transformer oil temperature prediction method

Abstract

The application provides a transformer oil temperature prediction model training method and a transformer oil temperature prediction method. It relates to the technical field of power safety. The method comprises the following steps: obtaining multiple groups of state data of a transformer based on state data of the transformer at different time points in a historical database; analyzing the influence degree of each feature on transformer oil temperature prediction based on the multiple groups of state data to obtain the influence intensity corresponding to each feature; obtaining target features associated with transformer oil temperature from multiple features according to the influence intensity corresponding to each feature and a preset threshold rule; selecting data corresponding to the target features in each group of state data to obtain multiple groups of training data; initializing a training model based on the multiple groups of training data and a parameter optimization algorithm; and training the training model based on the multiple groups of training data to obtain an oil temperature prediction model meeting training requirements. The application achieves the effect of timely and accurate oil temperature prediction.

Description

Technical Field

[0001] This application relates to the field of power safety technology, and in particular to a training method for an oil temperature prediction model and a method for predicting transformer oil temperature. Background Technology

[0002] Transformers are critical equipment in power systems, and their stable operation directly affects the safe and economical operation of the power grid. As an important indicator of the transformer's thermal state, the top oil temperature is crucial for accurately predicting the transformer's heat dissipation performance, insulation aging, and for preventing faults.

[0003] In related technologies, to address the issue of the top-level oil temperature of ultra-high voltage transformers being affected by multiple factors such as weather, time, and load, path analysis is used to decouple the direct and indirect effects of each factor on the oil temperature. Based on the direct path coefficient, the Euclidean distance is weighted and optimized to calculate the correlation between the weather, time, and load on the prediction day and the sample day. Finally, the similar time is selected and the top-level oil temperature is predicted. However, this approach suffers from low prediction accuracy and poor timeliness.

[0004] Therefore, there is an urgent need for a solution to accurately and timely predict transformer oil temperature in complex operating environments. Summary of the Invention

[0005] This application provides a training method for an oil temperature prediction model and a method for predicting transformer oil temperature, in order to achieve timely and accurate oil temperature prediction.

[0006] Firstly, this application provides a training method for an oil temperature prediction model, comprising:

[0007] Based on the transformer's state data at different times in the historical database, multiple sets of transformer state data are obtained, including data corresponding to multiple features.

[0008] Based on multiple sets of state data, the influence of each feature on transformer oil temperature prediction is analyzed to obtain the influence intensity of each feature.

[0009] Based on the influence intensity of each feature and the preset threshold rules, the target feature associated with transformer oil temperature is obtained from multiple features.

[0010] In each set of state data, select the data corresponding to the target feature to obtain multiple sets of training data;

[0011] The training model is initialized based on multiple sets of training data and parameter optimization algorithms;

[0012] Based on multiple sets of training data, the training model is trained to obtain an oil temperature prediction model that meets the training requirements. The oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

[0013] In one possible implementation, based on multiple sets of state data, the influence of each feature on transformer oil temperature prediction is analyzed to obtain the influence intensity corresponding to each feature, including:

[0014] Each feature in multiple sets of state data is taken as the feature to be processed in turn;

[0015] Add positive perturbations to the features to be processed in each set of state data to obtain positive perturbation data and , respectively; add negative perturbations to the features to be processed in each set of state data to obtain negative perturbation data and , respectively.

[0016] The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the positive perturbation prediction result corresponding to each set of state data.

[0017] The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the negative perturbation prediction result corresponding to each set of state data.

[0018] The influence intensity of the feature to be processed is obtained based on the difference between the positive and negative perturbation prediction results corresponding to each set of state data.

[0019] In one possible implementation, the parameter optimization algorithm is a particle swarm optimization algorithm. Based on multiple sets of training data and the parameter optimization algorithm, the training model is initialized, including:

[0020] Based on the network structure corresponding to the training model, multiple particles that meet the population size are initialized; where each particle is used to indicate a set of weights and thresholds within the training model.

[0021] Using test data, the network prediction error corresponding to each particle is determined, and the fitness corresponding to each particle is obtained. The test data is a number of training data sets randomly selected from multiple training data sets.

[0022] Based on the number of iterations of the particle swarm optimization algorithm and the fitness of each particle, determine whether the termination condition has been met.

[0023] When the termination condition is met, the training model is initialized based on the particle with the highest fitness.

[0024] If the termination condition is not met, the other particles are updated based on the particle with the highest fitness, resulting in multiple updated particles. The fitness of each updated particle is then redefined until the termination condition is met.

[0025] In one possible implementation, the training model is trained based on multiple sets of training data to obtain an oil temperature prediction model that meets the training requirements. This oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time period, and includes:

[0026] During the training process of the training model, the weights and thresholds are adjusted using the gradient descent method until the training model meets the training requirements, thus obtaining the oil temperature prediction model.

[0027] In one possible implementation, the training requirements include: the prediction error of the training model is lower than a preset error threshold, or the training epochs of the training model reach a preset maximum number of training epochs.

[0028] Secondly, this application provides a method for predicting transformer oil temperature, including:

[0029] Real-time acquisition of target parameters corresponding to the target features of the transformer, where the target features are those related to the transformer oil temperature among multiple features;

[0030] The target features are used as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the oil temperature prediction model training method of any of the various possible implementations of the first aspect above.

[0031] Thirdly, this application provides a training device for an oil temperature prediction model, comprising:

[0032] The acquisition module is used to obtain multiple sets of state data of the transformer based on the state data of the transformer at different times in the historical database. The state data includes data corresponding to multiple features.

[0033] The processing module is used to analyze the degree of influence of each feature on transformer oil temperature prediction based on multiple sets of state data, and obtain the influence intensity corresponding to each feature;

[0034] Based on the influence intensity of each feature and the preset threshold rules, the target feature associated with transformer oil temperature is obtained from multiple features; the data corresponding to the target feature is selected in each set of state data to obtain multiple sets of training data; the training model is initialized based on the multiple sets of training data and parameter optimization algorithm; the training model is trained based on the multiple sets of training data to obtain an oil temperature prediction model that meets the training requirements. The oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

[0035] In one possible implementation, the processing module is further configured to:

[0036] Each feature in multiple sets of state data is taken as the feature to be processed in turn;

[0037] Add positive perturbations to the features to be processed in each set of state data to obtain positive perturbation data and , respectively; add negative perturbations to the features to be processed in each set of state data to obtain negative perturbation data and , respectively.

[0038] The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the positive perturbation prediction result corresponding to each set of state data.

[0039] The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the negative perturbation prediction result corresponding to each set of state data.

[0040] The influence intensity of the feature to be processed is obtained based on the difference between the positive and negative perturbation prediction results corresponding to each set of state data.

[0041] In one possible implementation, the parameter optimization algorithm is a particle swarm optimization algorithm, and the processing module is specifically used for:

[0042] Based on the network structure corresponding to the training model, multiple particles that meet the population size are initialized; where each particle is used to indicate a set of weights and thresholds within the training model.

[0043] Using test data, the network prediction error corresponding to each particle is determined, and the fitness corresponding to each particle is obtained. The test data is a number of training data sets randomly selected from multiple training data sets.

[0044] Based on the number of iterations of the particle swarm optimization algorithm and the fitness of each particle, determine whether the termination condition has been met.

[0045] When the termination condition is met, the training model is initialized based on the particle with the highest fitness.

[0046] If the termination condition is not met, the other particles are updated based on the particle with the highest fitness, resulting in multiple updated particles. The fitness of each updated particle is then redefined until the termination condition is met.

[0047] In one possible implementation, the processing module is specifically used for:

[0048] During the training process of the training model, the weights and thresholds are adjusted using the gradient descent method until the training model meets the training requirements, thus obtaining the oil temperature prediction model.

[0049] In one possible implementation, the training requirements include: the prediction error of the training model is lower than a preset error threshold, or the training epochs of the training model reach a preset maximum number of training epochs.

[0050] Fourthly, this application provides a transformer oil temperature prediction device, comprising:

[0051] The acquisition module is used to acquire the target parameters corresponding to the target features of the transformer in real time. The target features are the features related to the transformer oil temperature among multiple features.

[0052] The processing module is used to take the target features as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the oil temperature prediction model training method of any of the various possible implementations of the first aspect above.

[0053] Fifthly, this application provides an electronic device, including: a memory and a processor;

[0054] The memory stores instructions that the computer executes;

[0055] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect, or the second aspect and / or various possible implementations of the second aspect.

[0056] In a sixth aspect, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect, or the second aspect and / or various possible implementations of the second aspect.

[0057] In a seventh aspect, this application provides a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect, or the second aspect and / or various possible implementations of the second aspect.

[0058] The oil temperature prediction model training method and transformer oil temperature prediction method provided in this application obtain multiple sets of transformer state data based on the transformer's state data at different times in a historical database. This helps to learn the complex relationship between the transformer's operating state and oil temperature, thereby improving the accuracy and generalization ability of the oil temperature prediction model. Selecting data corresponding to the target feature in each set of state data yields multiple sets of training data. This allows the oil temperature prediction model to focus on the target features that significantly influence oil temperature prediction, reducing interference from irrelevant data and improving the training efficiency and prediction accuracy of the model. Initializing the training model based on multiple sets of training data and a parameter optimization algorithm provides a reasonable starting point for parameters, helping the model converge to the optimal solution more quickly during training, thus improving training efficiency and model performance. Training the model based on multiple sets of training data yields an oil temperature prediction model that meets the training requirements, ensuring that the obtained model can accurately and quickly predict the transformer's oil temperature within a preset time period. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] Figure 1 A schematic diagram illustrating a training method for the oil temperature prediction model provided in this application embodiment;

[0061] Figure 2 A flowchart illustrating the training method for the oil temperature prediction model provided in this application embodiment. Figure 1 ;

[0062] Figure 3 A flowchart illustrating the training method for the oil temperature prediction model provided in this application embodiment. Figure 2 ;

[0063] Figure 4 A flowchart illustrating the transformer oil temperature prediction method provided in this application embodiment. Figure 1 ;

[0064] Figure 5 A schematic diagram of the structure of the training device for the oil temperature prediction model provided in the embodiments of this application;

[0065] Figure 6 This is a schematic diagram of the structure of the transformer oil temperature prediction device provided in the embodiments of this application;

[0066] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0067] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0068] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0069] In related technologies, to address the issue of the top-level oil temperature of ultra-high voltage transformers being affected by multiple factors such as weather, time, and load, path analysis is used to decouple the direct and indirect effects of each factor on the oil temperature. Based on the direct path coefficient, the Euclidean distance is weighted and optimized to calculate the correlation between the weather, time, and load on the prediction day and the sample day. Finally, the similar time is selected and the top-level oil temperature is predicted. However, this approach suffers from low prediction accuracy and poor timeliness.

[0070] The oil temperature prediction model training method provided in this application obtains multiple sets of training data by selecting data corresponding to the target feature in each set of state data. This allows the oil temperature prediction model to focus on the target feature that has a significant impact on oil temperature prediction, reducing interference from irrelevant data and improving the training efficiency and prediction accuracy of the oil temperature prediction model. Based on multiple sets of training data and parameter optimization algorithms, the training model is initialized, providing a reasonable starting point for parameters. This helps the training model converge to the optimal solution more quickly during training, improving training efficiency and performance. Based on multiple sets of training data, the training model is trained to obtain an oil temperature prediction model that meets the training requirements, ensuring that the obtained oil temperature prediction model can accurately and quickly predict the oil temperature of the transformer within a preset time.

[0071] Figure 1 This is a schematic diagram illustrating a scenario for training the oil temperature prediction model provided in an embodiment of this application. Figure 1 As shown, the specific application scenarios of this application include data center 11, processing center 12, transformer 13, and sensor 14, wherein:

[0072] Sensor 14 is deployed inside and outside transformer 13 and is used to collect status data of transformer 13. Sensor 14 is communicatively connected to data center 11 and can send the collected status data of transformer 13 to data center 11.

[0073] Data center 11 is communicatively connected to processing center 12 and sensor 14. Data center 11 can receive status data sent by sensor 14 and store this status data in chronological order to form a historical dataset containing transformer status data.

[0074] Processing center 12 can obtain multiple sets of transformer status data based on the transformer's status data at different times in the historical database, through communication connection with data center 11. Then, based on these multiple sets of status data, an oil temperature prediction model is trained.

[0075] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0076] Figure 2 A flowchart illustrating the training method for the oil temperature prediction model provided in this application embodiment. Figure 1 .like Figure 2 As shown, the method includes:

[0077] S201. Based on the state data of the transformer at different times in the historical database, obtain multiple sets of state data of the transformer, wherein the state data includes data corresponding to multiple features.

[0078] The historical database stores transformer status data at different times. By analyzing the data corresponding to multiple features within the status data, specific data values ​​or classification values ​​corresponding to various features of the transformer can be obtained, reflecting various indicators of the transformer's operating status. Optionally, these multiple features may include at least one of the following: transformer load current, ambient temperature, cooling system status, etc.

[0079] By retrieving transformer state data from historical databases at different times, multiple sets of transformer state data are obtained. Acquiring these multiple sets of transformer state data helps to learn the complex relationship between transformer operating state and oil temperature, thereby improving the accuracy and generalization ability of the oil temperature prediction model.

[0080] For example, a power system's historical database stores operational data of internal transformers. This historical database records the status data of multiple transformers every 15 minutes over the past year. By writing database query statements, multiple sets of status data at different times are extracted from the historical database in chronological order, thus obtaining multiple sets of transformer status data.

[0081] S202. Based on multiple sets of state data, analyze the degree of influence of each feature on transformer oil temperature prediction, and obtain the influence intensity corresponding to each feature.

[0082] The degree of influence indicates the magnitude of each feature's impact on transformer oil temperature prediction. The strength of influence is a quantitative representation of this degree of influence. Optionally, the strength of influence can be calculated using specific data analysis methods to measure the closeness of the correlation between each feature and transformer oil temperature prediction.

[0083] Optionally, the data analysis method includes any one of correlation analysis, regression analysis, or feature importance assessment methods in machine learning. Specifically, correlation analysis includes any one of the following: Pearson correlation coefficient and Spearman correlation coefficient; regression analysis includes any one of the following: linear regression and logistic regression; and feature importance assessment methods include any one of the following: feature importance based on decision trees and feature importance based on random forests.

[0084] Data analysis methods are used to analyze the relationship between each feature in multiple sets of state data and transformer oil temperature prediction, and the influence strength of each feature is calculated. By quantifying the influence strength of each feature on transformer oil temperature prediction, the degree of correlation between each feature and transformer oil temperature prediction can be clearly understood. This helps to screen out target features that are closely related to transformer oil temperature prediction, thereby improving the accuracy and efficiency of the oil temperature prediction model.

[0085] S203. Based on the influence intensity of each feature and the preset threshold rules, the target feature associated with transformer oil temperature is obtained from multiple features.

[0086] The target features are those closely related to transformer oil temperature in the state data. These target features play a crucial role in the variation of oil temperature. Target features are selected from each set of state data to form training data for training the oil temperature prediction model.

[0087] The preset threshold rules are pre-defined standards used to determine the correlation between features and transformer oil temperature. These preset threshold rules can be determined based on actual needs, experience, or experiments. For example, a threshold value for the intensity of influence is set as the preset threshold rule. When the influence intensity of a feature exceeds the threshold, it is considered that the feature has a strong influence on transformer oil temperature prediction.

[0088] Based on the influence strength of each feature, a preset threshold rule is used for filtering. If the influence strength of a feature meets the threshold rule, then that feature is identified as the target feature; otherwise, it is excluded. By filtering each feature among multiple features, the target features that have a significant impact on transformer oil temperature prediction can be accurately selected from multiple features, while irrelevant or less influential features are removed. This reduces the complexity of the training model, improves the training efficiency and prediction accuracy of the oil temperature prediction model, and reduces the risk of overfitting.

[0089] S204. Select the data corresponding to the target feature in each set of state data to obtain multiple sets of training data.

[0090] Specifically, the data corresponding to the target feature is selected from each set of state data. Then, the data corresponding to the target feature are combined to form multiple sets of training data.

[0091] For example, if the state data includes load current, ambient temperature, and cooling system status, and the target features are load current and ambient temperature, then the data corresponding to the load current and ambient temperature are sequentially extracted from each set of state data to form multiple sets of training data.

[0092] By selecting data corresponding to target features associated with oil temperature as training data, the oil temperature prediction model can focus on factors that have a significant impact on oil temperature prediction, reduce interference from irrelevant data, and improve the training efficiency and prediction accuracy of the oil temperature prediction model.

[0093] Optionally, the target feature can be determined from multiple features through expert analysis and data correlation tests.

[0094] Optionally, feature selection algorithms in machine learning can be used to select the target feature most relevant to oil temperature from multiple features.

[0095] S205. Initialize the training model based on multiple sets of training data and parameter optimization algorithms.

[0096] A parameter optimization algorithm is used to adjust the parameters of the trained model. Optionally, the parameter optimization algorithm can be at least one of gradient descent, genetic algorithm, etc. The trained model is an initial model structure used to predict transformer oil temperature that has not yet been fully trained. Optionally, the trained model is built based on a neural network model.

[0097] By using parameter optimization algorithms, the initial parameters of the training model are set and adjusted based on multiple sets of training data, providing a reasonable starting point for the training model. This helps the training model converge to the optimal solution more quickly during the training process, thereby improving training efficiency and the performance of the training model.

[0098] For example, using gradient descent, the gradient of the loss function with respect to the parameters is calculated based on the set initial parameters and learning rate. The parameters are then updated in the opposite direction of the gradient to obtain the updated parameters. The training model is then initialized based on the updated parameters.

[0099] Optionally, gradient descent can be chosen as the parameter optimization algorithm to initialize a neural network-based model to obtain a training model. First, initial weights and bias parameters are set for each layer of the neural network. Then, based on gradient descent, the gradient of the loss function with respect to the initial weights and bias parameters is calculated using the training data, yielding the first gradient. Afterward, based on the first gradient, the parameters are updated according to the rules of gradient descent, completing the initialization of the training model.

[0100] Optionally, the training model is a backpropagation (BP) neural network. A gradient descent strategy is used to minimize the error function between the network output and the expected value. The chain rule is then used to propagate the error signal from the output layer back along the network topology, thereby iteratively optimizing the weight matrix and threshold parameters of the BP neural network. Based on the parameter adjustment mechanism along the error gradient direction, the BP neural network can gradually approximate the objective function, ultimately achieving high-precision nonlinear data fitting. The objective function is the minimization of the loss function. The convergence of the BP neural network depends on the learning rate, initial parameter settings, and a reasonable configuration of the network structure.

[0101] For example, in a BP network, n sets of training data are input, where each set of training data includes p features, which can be represented as follows: After inputting n sets of training data, the predicted oil temperature is obtained as the output. Where j = 1, 2, ..., m. Then, a squared error function is used to obtain the training error of the nth set of training data. for:

[0102]

[0103] in, This indicates the actual oil temperature.

[0104] Subsequently, during the training of the BP network, gradient descent was used to change the weights of the BP network, and the gradient error between neuron i and neuron j was obtained. for:

[0105]

[0106] in, Indicates the learning rate; This represents the training error of the nth set of training data; This represents the connection weight between neuron i and neuron j.

[0107] Furthermore, due to ,in, This represents the output layer corresponding to neuron i; This represents the error signal of neuron j, and the gradient error between neuron i and neuron j is obtained. for:

[0108]

[0109] If neuron j is an output layer neuron, then the error signal of neuron j... It can mean "to think":

[0110]

[0111] in, This represents the expected output of neuron j; This represents the predicted output of neuron j.

[0112] If neuron j is a hidden layer neuron, then the error signal of neuron j... It can mean "to think":

[0113]

[0114] in, This represents the connection weights from hidden layer neuron h to the next layer neuron j. This represents the output of the hidden layer neuron h.

[0115] Then, based on the weight value expression, the weight values ​​of the BP neural network in the m-th training process are obtained as follows:

[0116]

[0117] in, : Represents the connection weights from neuron i to neuron j during the m-th training iteration; η represents the weight update amount between neurons i and j during the (m-1)th training iteration; η represents the learning rate. This represents the error signal of neuron j; This represents the output of neuron i in the (n-1)th layer.

[0118] S206. Based on multiple sets of training data, the training model is trained to obtain an oil temperature prediction model that meets the training requirements. The oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

[0119] Training requirements typically include specifications for the oil temperature prediction model, such as prediction accuracy and error range. The training model is considered complete only when its performance on the training data meets the requirements, and this model is then used as the oil temperature prediction model. The oil temperature prediction model is used to predict the oil temperature of a transformer within a preset time period.

[0120] Multiple sets of training data are input into the initialized training model. The training model calculates and predicts based on each set of training data, obtaining a prediction result for each set. Then, the prediction result for each set of training data is compared with the actual oil temperature data, and a loss function is calculated. Next, the parameters of the training model are adjusted based on the value of the loss function, and predictions are re-performed for each set of training data until the prediction results of the training model meet the training requirements, resulting in a satisfactory oil temperature prediction model.

[0121] By training the model with multiple sets of training data, the model can learn the inherent relationship between transformer state data and oil temperature, ensuring that the resulting oil temperature prediction model can accurately predict the transformer oil temperature within a preset time, providing an important reference for the operation and maintenance of the power system.

[0122] Optionally, multiple sets of training data can be input into a training model based on a Long Short-Term Memory (LSTM) network. The training model performs forward propagation calculations based on each set of input training data, outputting a predicted oil temperature. The predicted oil temperature is compared with the actual oil temperature data, and the mean squared error is calculated as the loss function. Simultaneously, a training requirement is set that the loss function is less than a loss threshold. If the loss function is greater than the loss threshold, the model's weight parameters are adjusted using the backpropagation algorithm, and training is re-run. After multiple iterations of training, when the loss function is less than the loss threshold, an oil temperature prediction model that meets the training requirements is obtained.

[0123] The oil temperature prediction model training method provided in this application obtains multiple sets of transformer state data based on the transformer's state data at different times in a historical database. This helps learn the complex relationship between the transformer's operating state and oil temperature, thereby improving the accuracy and generalization ability of the oil temperature prediction model. Selecting data corresponding to the target feature in each set of state data yields multiple sets of training data. This allows the oil temperature prediction model to focus on the target features that significantly influence oil temperature prediction, reducing interference from irrelevant data and improving the training efficiency and prediction accuracy of the model. Initializing the training model based on multiple sets of training data and a parameter optimization algorithm provides a reasonable starting point for parameters, helping the model converge to the optimal solution more quickly during training, thus improving training efficiency and model performance. Training the model based on multiple sets of training data yields an oil temperature prediction model that meets the training requirements, ensuring that the obtained model can accurately and quickly predict the transformer's oil temperature within a preset time period.

[0124] Figure 3 A flowchart illustrating the training method for the oil temperature prediction model provided in this application embodiment. Figure 2 .like Figure 3 As shown, in this embodiment... Figure 2 Based on the examples, the training method for the oil temperature prediction model is described in detail. The method includes:

[0125] For example, step S202 above may further include:

[0126] S2021. Sequentially treat each feature in multiple sets of state data as a feature to be processed.

[0127] By iterating through all features in multiple sets of state data and selecting one feature at a time as the feature to be processed in the current analysis, each feature can be comprehensively analyzed, ensuring that no feature that may affect the transformer oil temperature is missed.

[0128] S2022. Add positive perturbations to the features to be processed in each set of state data to obtain positive perturbation data corresponding to multiple sets of state data, and add negative perturbations to the features to be processed in each set of state data to obtain negative perturbation data corresponding to multiple sets of state data.

[0129] For each set of state data, after determining the feature to be processed, a positive perturbation value is added to the feature to obtain positive perturbation data, and a negative perturbation value is added to obtain negative perturbation data, according to a preset perturbation amplitude. Other features in each set of state data remain unchanged. Optionally, the perturbation amplitude can be 5% of the feature value, 10% of the feature value, or any proportion of the feature value.

[0130] By adding positive and negative perturbations, the state of the feature to be processed under different changing conditions can be simulated, thereby observing the impact of the feature to be processed on the transformer oil temperature prediction results.

[0131] S2023. Use the positive perturbation data corresponding to each set of state data as input to the pre-trained model to obtain the positive perturbation prediction result corresponding to each set of state data.

[0132] The positive disturbance data corresponding to each set of state data is sequentially input into the pre-trained model, and the model is run to make predictions, so as to obtain the positive disturbance prediction results of transformer oil temperature after adding positive disturbances to the features to be processed.

[0133] Optionally, the pre-trained model can be an existing pre-trained oil temperature prediction model; it can also be an oil temperature prediction model trained by selecting a small sample from the training data; furthermore, the pre-trained model can be an oil temperature prediction model obtained in this embodiment and any possible implementation thereof.

[0134] S2024. Use the positive perturbation data corresponding to each set of state data as input to the pre-trained model to obtain the negative perturbation prediction result corresponding to each set of state data.

[0135] The negative perturbation data corresponding to each set of state data is sequentially input into the pre-trained model, the model is run to make predictions, and the oil temperature prediction value corresponding to each set of negative perturbation data is obtained, that is, the negative perturbation prediction result.

[0136] S2025. Based on the difference between the positive and negative disturbance prediction results corresponding to each set of state data, the influence intensity of the feature to be processed is obtained.

[0137] The difference between the positive and negative disturbance prediction results reflects the magnitude of the change in oil temperature prediction results under positive and negative disturbance conditions. The absolute value of the difference between the positive and negative disturbance prediction results corresponding to each set of state data is taken to obtain the influence intensity of the feature to be processed.

[0138] When the difference between the positive and negative disturbance prediction results for each set of state data is negative, the influence of the feature to be processed on the transformer oil temperature is suppressive. When the difference between the positive and negative disturbance prediction results for each set of state data is positive, the influence of the feature to be processed on the transformer oil temperature is facilitative.

[0139] For each set of state data, the difference between the corresponding positive and negative disturbance prediction results is calculated to obtain the disturbance difference for each set of state data. Then, statistical analysis is performed based on the disturbance differences corresponding to multiple sets of state data to obtain the influence intensity of the feature to be processed. By combining the negative and positive disturbance prediction results, the impact of the feature to be processed on transformer oil temperature prediction can be analyzed more comprehensively.

[0140] In one possible implementation, the parameter optimization algorithm is a particle swarm optimization algorithm, and step S205 may further include:

[0141] S2051. Based on the network structure corresponding to the training model, initialize multiple particles that meet the population size; wherein, each particle is used to indicate a set of weights and thresholds within the training model.

[0142] Population size describes the number of particles. Population size determines the number of solutions the particle swarm optimization (PSO) algorithm can explore simultaneously in the search space. A larger population size results in stronger search capabilities, but also increases computational complexity. A particle is the basic unit in the PSO algorithm; each particle represents a set of weights and thresholds in the training model.

[0143] By training the network structure corresponding to the model, the number of weights and thresholds for the training model are determined. Based on the number of weights and thresholds, multiple particles equal to the population size are randomly generated, such that each particle can indicate a set of weights and thresholds within the training model.

[0144] Optionally, the number of weights for the training model can be determined by obtaining the number of neurons in different layers; the number of thresholds for the training model can be determined based on the connection method between neurons in different layers. By initializing multiple particles, the particle swarm optimization algorithm can explore multiple solutions in the search space simultaneously, increasing the probability of finding the global optimum and helping to improve the performance of the training model.

[0145] For example, if the number of weights is 4 and the number of thresholds is 2, then each particle should include 4 weights and 2 thresholds.

[0146] S2052. Using test data, determine the network prediction error corresponding to each particle and obtain the fitness corresponding to each particle. The test data is a number of training data sets randomly selected from multiple training data sets.

[0147] Test data is a randomly selected portion of the training data used to evaluate the performance of the neural network model corresponding to the particle. Network prediction error is the difference between the predicted oil temperature and the actual oil temperature when the neural network model predicts the test data. Fitness is used to measure the particle's performance. Generally, the smaller the network prediction error, the higher the particle's fitness, indicating a better combination of weights and thresholds for that particle.

[0148] The weights and thresholds corresponding to each particle are assigned to the training model to obtain the training model for each particle. Then, the training model for each particle is used to predict the oil temperature using test data, resulting in the predicted oil temperature. The fitness of each particle is obtained by measuring the network prediction error between the predicted oil temperature and the actual oil temperature, thus evaluating the performance of each particle and guiding the algorithm towards a better solution.

[0149] S2053. Based on the number of iterations of the particle swarm optimization algorithm and the fitness of each particle, determine whether the termination condition has been met.

[0150] After each iteration, the termination condition is determined based on the number of iterations of the particle swarm optimization algorithm and the fitness of each particle.

[0151] The termination condition is the condition under which the particle swarm optimization algorithm stops searching. Optional termination conditions include whether the number of iterations has reached a preset maximum number of iterations, or whether the fitness is greater than a preset fitness value.

[0152] For example, after each iteration, it is checked whether the current iteration count has reached the preset maximum iteration count, or whether the fitness is greater than the preset fitness value. If either condition is met, the termination condition is determined to have been met.

[0153] S2054. When the termination condition is met, the training model is initialized based on the particle with the highest fitness.

[0154] When the termination condition is reached, the weights and thresholds of the particle with the highest fitness are used as the globally optimal weight and threshold combination. The training model is then initialized based on this globally optimal combination. Initializing the training model using the particle with the highest fitness ensures good performance from the start of training or application, thus improving the convergence speed and final performance of the training model.

[0155] S2055. When the termination condition is not met, update other particles based on the particle with the highest fitness to obtain multiple updated particles, and redetermine the fitness of each updated particle until the termination condition is met.

[0156] For each particle, based on its current velocity, its historical best position, and the position of the globally best particle, new velocity and position are calculated according to the particle swarm optimization (PSO) velocity and position update formulas. Then, the fitness is recalculated using the updated particles. This process is repeated until the termination condition is met. By continuously updating particles, the PSO algorithm can gradually move towards better solutions in the search space, increasing the probability of finding the globally optimal solution and thus optimizing the performance of the trained model.

[0157] Furthermore, when updating particles, a mutation operator is introduced to enhance particle diversity and avoid premature convergence.

[0158] Optionally, the velocity update formula for the particle swarm optimization algorithm can be expressed as:

[0159]

[0160] in, Indicates in The velocity of the i-th particle in the d-th dimension during the next iteration; Indicates in The velocity of the i-th particle in the d-th dimension during the next iteration; W is the inertia weight; C1 and C2 are learning factors; R1 and R2 are random numbers in the interval [0, 1]. Indicates in The particle's best historical position in d-dimensional dimension at the next iteration; exist The particle's global optimal position in the d-dimensional population at the next iteration; The position of the i-th particle in the d-th dimension at the next iteration.

[0161] The velocity update formula and position update formula of the particle swarm optimization algorithm can be expressed as follows:

[0162]

[0163] in, Indicates in The position of the i-th particle in the d-th dimension at the next iteration; The position of the i-th particle in the d-th dimension at the next iteration; Indicates in The velocity of the i-th particle in the d-th dimension during the next iteration.

[0164] In one possible implementation, step S206 may further include: during the training process of the training model, adjusting the weights and thresholds using gradient descent until the training model meets the training requirements, thereby obtaining the oil temperature prediction model.

[0165] Gradient descent is a method used to optimize a loss function. It calculates the gradient of the loss function with respect to weights and thresholds, and updates the weights and thresholds in the opposite direction of the gradient to gradually reduce the value of the loss function, thereby making the prediction results of the trained model closer to the true value.

[0166] Weights are the strength of connections between different neurons in a training model; they determine the degree to which the input signal influences the output signal. Thresholds are parameters within neurons in a training model, used to control the activation state of neurons; neurons are only activated when the weighted input exceeds the threshold.

[0167] During model training, firstly, the loss between the current model's predictions on the training data and the actual results is calculated. Then, the gradient of the loss function with respect to each weight and threshold is calculated. Based on the gradient and a preset learning rate, the weights and thresholds are updated in the opposite direction of the gradient. This process is repeated, continuously adjusting the weights and thresholds until the training requirements are met; the resulting model is the oil temperature prediction model. By adjusting the weights and thresholds using gradient descent, the training model can gradually learn the mapping relationship between the input data and the actual oil temperature, improving the prediction accuracy of the oil temperature prediction model.

[0168] In one possible implementation, the training requirements include: the prediction error of the training model is lower than a preset error threshold, or the training epochs of the training model reach a preset maximum number of training epochs.

[0169] Prediction error is used to indicate the difference between the predicted result and the true result when the trained model makes predictions on training data. Optionally, prediction error includes at least one of mean square error, mean absolute error and root mean square error.

[0170] Specifically, the mean squared error can be expressed as:

[0171]

[0172] in, represents the mean squared error, and N represents the number of training data sets; Represents the i-th set of training data The predicted oil temperature value This represents the actual oil temperature of the i-th training data set.

[0173] The mean absolute error can be expressed as:

[0174]

[0175] in, represents the mean absolute error, and N represents the number of training data sets; This represents the predicted oil temperature value for the i-th training data set. This represents the actual oil temperature of the i-th training data set.

[0176] The root mean square error can be expressed as:

[0177]

[0178] in, denoted as root mean square error, and N represents the number of training data sets; This represents the predicted oil temperature value for the i-th training data set. This represents the actual oil temperature of the i-th training data set.

[0179] Training rounds refer to the number of forward and backward propagation processes performed on all training data during neural network training.

[0180] The preset maximum number of training rounds is the maximum number of training sessions set in advance. It is used to limit the time and computational resource consumption of the training process and prevent overtraining or infinite loops.

[0181] During model training, the prediction error of the current model is calculated after each training epoch. This prediction error is compared to a preset error threshold, and the number of training epochs completed is recorded. If the prediction error is below the preset threshold, or the maximum number of training epochs has been reached, training stops; otherwise, the next training epoch continues. Training conditions allow for flexible control of the training process. When the model's prediction error is sufficiently small, training stops promptly to avoid overfitting; when the maximum number of training epochs has been reached, training time is prevented from becoming excessively long, improving training efficiency.

[0182] Figure 4 This is a schematic flowchart illustrating the transformer oil temperature prediction method provided in an embodiment of this application. Figure 4 As shown, the method includes:

[0183] S401. Real-time acquisition of target parameters corresponding to the target features of the transformer. The target features are those related to the transformer oil temperature among multiple features.

[0184] Various sensors are deployed at the transformer operating site. These sensors collect the transformer's physical quantities. Then, based on the transformer's physical quantities and target characteristics, the corresponding target parameters are obtained, thus enabling real-time acquisition of these parameters. By acquiring these target parameters related to transformer oil temperature in real time, changes in various factors affecting the transformer oil temperature can be monitored promptly, helping to detect potential overheating and other abnormalities in advance, ensuring the safe and stable operation of the transformer.

[0185] Optionally, the various sensors can be at least one of the following: voltage sensor, current sensor, temperature sensor, wind speed sensor, and flow sensor.

[0186] For example, a current sensor measures the transformer's load current; a temperature sensor measures the transformer's ambient temperature; and a flow sensor monitors the flow rate of the coolant in the transformer's cooling system. These various sensors convert the real-time physical quantities into electrical signals, which are then converted into digital signals by a data acquisition system and transmitted to a data processing unit, thereby enabling real-time acquisition of target parameters corresponding to the transformer's target characteristics.

[0187] S402. The target features are used as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the training method of the oil temperature prediction model in the above embodiments and any possible implementation of the above embodiments.

[0188] The real-time acquired target feature parameters are organized and preprocessed according to the format required by the oil temperature prediction model. Then, the preprocessed target features are used as input and fed into the pre-trained oil temperature prediction model. The oil temperature prediction model calculates based on the internally learned mapping relationship and outputs the predicted oil temperature value of the transformer within a preset time period. It can quickly and accurately predict the oil temperature of the transformer within a preset time period, providing forward-looking information for the operation and maintenance of the transformer.

[0189] Figure 5 This is a schematic diagram of the structure of the training device for the oil temperature prediction model provided in an embodiment of this application. Figure 5 As shown, the training device 50 for the oil temperature prediction model provided in this embodiment includes:

[0190] The acquisition module 501 is used to obtain multiple sets of state data of the transformer based on the state data of the transformer at different times in the historical database. The state data includes data corresponding to multiple features.

[0191] The processing module 502 is used to analyze the influence of each feature on transformer oil temperature prediction based on multiple sets of state data, and obtain the influence intensity corresponding to each feature; according to the influence intensity corresponding to each feature and the preset threshold rule, the target feature associated with transformer oil temperature is obtained among multiple features; the data corresponding to the target feature is selected from each set of state data to obtain multiple sets of training data; the training model is initialized based on multiple sets of training data and parameter optimization algorithm; the training model is trained based on multiple sets of training data to obtain an oil temperature prediction model that meets the training requirements. The oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

[0192] In one possible implementation, the processing module 502 is further configured to: sequentially use each feature in multiple sets of state data as a feature to be processed; add a positive perturbation to the feature to be processed in each set of state data to obtain positive perturbation data and negative perturbation data corresponding to multiple sets of state data; use the positive perturbation data corresponding to each set of state data as input to a pre-trained model to obtain a positive perturbation prediction result corresponding to each set of state data; use the positive perturbation data corresponding to each set of state data as input to a pre-trained model to obtain a negative perturbation prediction result corresponding to each set of state data; and obtain the influence intensity of the feature to be processed based on the difference between the positive perturbation prediction result and the negative perturbation prediction result corresponding to each set of state data.

[0193] In one possible implementation, the parameter optimization algorithm is a particle swarm optimization algorithm. The processing module 502 is specifically used for: initializing multiple particles that meet the population size based on the network structure corresponding to the training model; wherein each particle is used to indicate the weights and thresholds within a set of training models; using test data, determining the network prediction error corresponding to each particle, and obtaining the fitness corresponding to each particle, wherein the test data is a number of training data sets randomly selected from multiple sets of training data; determining whether the termination condition has been reached based on the number of iterations of the particle swarm optimization algorithm and the fitness corresponding to each particle; when the termination condition is reached, initializing the training model based on the particle with the highest fitness; when the termination condition is not reached, updating other particles based on the particle with the highest fitness, obtaining multiple updated particles, and re-determining the fitness corresponding to each updated particle, until the termination condition is reached.

[0194] In one possible implementation, the processing module 502 is specifically used to: adjust the weights and thresholds using gradient descent during the training process of the training model until the training model meets the training requirements and obtains the oil temperature prediction model.

[0195] In one possible implementation, the training requirements include: the prediction error of the training model is lower than a preset error threshold, or the training epochs of the training model reach a preset maximum number of training epochs.

[0196] The training device for the oil temperature prediction model provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0197] Figure 6 This is a schematic diagram of the transformer oil temperature prediction device provided in an embodiment of this application. Figure 6 As shown, the transformer oil temperature prediction device 60 provided in this embodiment includes:

[0198] The acquisition module 601 is used to acquire the target parameters corresponding to the target features of the transformer in real time. The target features are the features related to the transformer oil temperature among multiple features.

[0199] The processing module 602 is used to take the target features as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the oil temperature prediction model training method of any of the various possible implementations of the first aspect above.

[0200] The transformer oil temperature prediction device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0201] Figure 7 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 7 As shown, the electronic device 70 provided in this embodiment includes at least one processor 701 and a memory 702. Optionally, the device 70 further includes a communication component 703. The processor 701, memory 702, and communication component 703 are connected via a bus 704.

[0202] In a specific implementation, at least one processor 701 executes computer execution instructions stored in memory 702, causing at least one processor 701 to perform the above-described method.

[0203] The specific implementation process of processor 701 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0204] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0205] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0206] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings of this application's embodiments are not limited to only one bus or one type of bus.

[0207] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0208] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, implement any of the methods described above.

[0209] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0210] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0211] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0212] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0213] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0214] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0215] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0216] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A training method for an oil temperature prediction model, characterized in that, include: Based on the transformer's state data at different times in the historical database, multiple sets of state data for the transformer are obtained, wherein the state data includes data corresponding to multiple features; Based on multiple sets of state data, the degree of influence of each feature on the transformer oil temperature prediction is analyzed to obtain the influence intensity corresponding to each feature. Based on the influence intensity corresponding to each feature and the preset threshold rule, the target feature associated with the transformer oil temperature is obtained from among the multiple features; In each set of state data, select the data corresponding to the target feature to obtain multiple sets of training data; The training model is initialized based on multiple sets of training data and parameter optimization algorithms. Based on multiple sets of training data, the training model is trained to obtain an oil temperature prediction model that meets the training requirements. The oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

2. The method according to claim 1, characterized in that, The analysis of the impact of each feature on the transformer oil temperature prediction based on multiple sets of state data yields the influence intensity corresponding to each feature, including: Each feature in the multiple sets of state data is sequentially taken as a feature to be processed; Add a positive perturbation to the feature to be processed in each set of state data to obtain positive perturbation data corresponding to multiple sets of state data, and add a negative perturbation to the feature to be processed in each set of state data to obtain negative perturbation data corresponding to multiple sets of state data; The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the positive perturbation prediction result corresponding to each set of state data. The positive perturbation data corresponding to each set of state data is used as the input to the pre-trained model to obtain the negative perturbation prediction result corresponding to each set of state data. The influence intensity of the feature to be processed is obtained based on the difference between the positive disturbance prediction result and the negative disturbance prediction result corresponding to each set of state data.

3. The method according to claim 1, characterized in that, The parameter optimization algorithm is a particle swarm optimization algorithm. The initialization of the training model based on multiple sets of training data and the parameter optimization algorithm includes: Based on the network structure corresponding to the training model, multiple particles that meet the population size are initialized; wherein, each particle is used to indicate a set of weights and thresholds within the training model; Using test data, the network prediction error corresponding to each particle is determined, and the fitness corresponding to each particle is obtained. The test data is a number of sets of training data randomly selected from multiple sets of training data. Based on the number of iterations of the particle swarm optimization algorithm and the fitness of each particle, determine whether the termination condition has been met. When the termination condition is met, the training model is initialized based on the particle with the highest fitness. If the termination condition is not met, other particles are updated based on the particle with the highest fitness, resulting in multiple updated particles. The fitness of each updated particle is then redefined until the termination condition is met.

4. The method according to claim 3, characterized in that, The training model is trained based on multiple sets of training data to obtain an oil temperature prediction model that meets the training requirements. This oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time period, including: During the training process of the training model, the weights and the threshold are adjusted using the gradient descent method until the training model meets the training requirements, thus obtaining the oil temperature prediction model.

5. The method according to claim 1, characterized in that, The training requirements include: the prediction error of the training model is lower than a preset error threshold, or the training epochs of the training model reach a preset maximum number of training epochs.

6. A method for predicting transformer oil temperature, characterized in that, include: Real-time acquisition of target parameters corresponding to target features of the transformer, wherein the target features are features related to transformer oil temperature among multiple features; The target feature is used as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the training method of the oil temperature prediction model according to any one of claims 1-5.

7. A training device for an oil temperature prediction model, characterized in that, include: The acquisition module is used to obtain multiple sets of state data of the transformer based on the state data of the transformer at different times in the historical database, wherein the state data includes data corresponding to multiple features; The processing module is used to analyze the influence of each feature on the transformer oil temperature prediction based on multiple sets of state data, and obtain the influence intensity corresponding to each feature; according to the influence intensity corresponding to each feature and a preset threshold rule, obtain the target feature associated with the transformer oil temperature among multiple features; select the data corresponding to the target feature in each set of state data to obtain multiple sets of training data; initialize the training model based on the multiple sets of training data and a parameter optimization algorithm; train the training model based on the multiple sets of training data to obtain an oil temperature prediction model that meets the training requirements, and the oil temperature prediction model is used to predict the oil temperature of the transformer within a preset time.

8. A training device for an oil temperature prediction model, characterized in that, include: The acquisition module is used to acquire the target parameters corresponding to the target features of the transformer in real time. The target features are the features related to the transformer oil temperature among multiple features. The processing module is used to take the target features as input to the oil temperature prediction model to obtain the oil temperature of the transformer within a preset time. The oil temperature prediction model is trained by the training method of the oil temperature prediction model according to any one of claims 1-5.

9. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1-6.

11. A computer program product, characterized in that, Includes a computer program, which, when executed, implements the method according to any one of claims 1-6.

Images