Construction method of transformer winding temperature prediction model, temperature prediction method, construction device, computer equipment and storage medium

Abstract

The invention relates to a construction method of a transformer winding temperature prediction model, a temperature prediction method, a construction device, computer equipment and a storage medium. Acquiring a plurality of winding temperature sample data; dividing the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets; determining a plurality of training sets and corresponding verification sets based on the plurality of data sets; training the plurality of convolutional neural network models based on the plurality of training sets to obtain a plurality of temperature prediction models; performing index evaluation on the plurality of temperature prediction models based on the plurality of verification sets to obtain a plurality of evaluation results; and determining a target prediction model based on the plurality of evaluation results. According to the method, the model performance can be remarkably improved, so that the temperature distribution of the transformer winding can be predicted more accurately.

Description

Technical Field

[0001] The present application relates to the technical field of winding temperature prediction, and in particular to a method for constructing a transformer winding temperature prediction model, a temperature prediction method, a construction device, a computer device and a storage medium. Background Art

[0002] With the continuous development of power technology, power transformers are increasingly used. The load capacity and usable life of power transformers also depend on whether the heat generated inside them can be transferred to the surrounding environment in a timely manner, that is, thermal characteristics. Therefore, it is of great significance to study the complex thermal process inside the transformer, establish a transformer thermal model, and conduct research on transformer condition maintenance based on this.

[0003] In traditional technology, transformer thermal models can be established through statistical modeling, machine learning and deep learning. Among them, deep learning can automatically extract high-level feature representations from massive temperature field data. These features not only include the spatial and temporal characteristics of temperature distribution, but also imply the complex relationship between temperature and various influencing factors. Through multi-layer nonlinear transformation, deep learning models can capture these potential laws, thereby achieving accurate prediction of temperature fields. During the training process, the model gradually adapts to the changes in temperature fields under different working conditions and conditions by continuously learning and adjusting parameters. Therefore, when faced with new input data, the model can quickly make accurate predictions without additional adjustments or training. However, in traditional deep learning methods, such as BP (Back Propagation) neural networks and convolutional neural networks, although they can automatically extract high-level features and achieve accurate predictions, they often focus on the direct relationship between various parts of the input data and the output when extracting features, and it is difficult to directly capture global features at different scales, resulting in inaccurate predictions of the temperature distribution of transformer windings. Summary of the invention

[0004] Based on this, it is necessary to provide a method for constructing a transformer winding temperature prediction model, a temperature prediction method, a construction device, a computer device and a storage medium that can more accurately predict the temperature distribution of the transformer winding in response to the above technical problems.

[0005] In the first aspect, the present application provides a method for constructing a transformer winding temperature prediction model. The method includes: obtaining multiple winding temperature sample data; wherein the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution; dividing the multiple winding temperature sample data into multiple non-overlapping subsets to obtain multiple data sets; determining multiple training sets and corresponding validation sets based on the multiple data sets; wherein each of the data sets is used as the validation set; training multiple convolutional neural network models based on the multiple training sets to obtain multiple temperature prediction models; wherein the convolutional neural network model includes multiple parallel convolutional layers with different convolution kernels; performing index evaluation on multiple temperature prediction models based on the multiple validation sets to obtain multiple evaluation results; determining a target prediction model based on the multiple evaluation results.

[0006] In one embodiment, a plurality of original winding temperature data are acquired; and the plurality of original winding temperature data are standardized to obtain a plurality of winding temperature sample data.

[0007] In one of the embodiments, the step of training multiple convolutional neural network models based on the multiple training sets includes: inputting all the winding temperature sample data in the training set into the convolutional neural network model to obtain a predicted temperature distribution; updating the model parameters of the convolutional neural network model based on the predicted temperature distribution and the winding temperature distribution to obtain the temperature prediction model.

[0008] In one of the embodiments, the step of inputting all the winding temperature sample data in the training set into the convolutional neural network model to obtain the predicted temperature distribution includes: performing convolution processing and activation processing on all the winding temperature sample data to obtain multiple convolution data; performing pooling processing on the multiple convolution data to obtain multiple pooling data; performing fusion processing on the multiple pooling data to obtain fusion data; and determining the predicted temperature distribution based on the fusion data.

[0009] In one of the embodiments, the step of determining the predicted temperature distribution based on the fused data includes: performing a first full-connection processing on the fused data to obtain first fully-connected data; performing a random data discarding processing on the first fully-connected data to obtain second fully-connected data; performing a second full-connection processing on the second fully-connected data to obtain the predicted temperature distribution.

[0010] In one of the embodiments, the step of training multiple convolutional neural network models based on the multiple training sets respectively also includes: determining a learning rate through a cosine function; and determining a speed and direction of updating model parameters of the convolutional neural network model based on the learning rate.

[0011] In the second aspect, the present application also provides a transformer winding temperature prediction method. The method comprises: obtaining current transformer state data; wherein the current transformer state data comprises: inlet pressure, inlet temperature, outlet pressure, outlet temperature and heat source parameters; inputting the current transformer state data into a target prediction model to obtain the current winding temperature distribution; wherein the target prediction model is obtained by the method for constructing the transformer winding temperature prediction model described in the first aspect embodiment.

[0012] In the third aspect, the present application also provides a device for constructing a transformer winding temperature prediction model. The device includes: a data acquisition module, which is used to acquire multiple winding temperature sample data; wherein the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution; a data division module, which is used to divide the multiple winding temperature sample data into multiple non-overlapping subsets to obtain multiple data sets; a data classification module, which is used to determine multiple training sets and corresponding validation sets based on the multiple data sets; wherein each of the data sets is used as the validation set; a model training module, which is used to train multiple convolutional neural network models based on the multiple training sets to obtain multiple temperature prediction models; wherein the convolutional neural network model includes multiple convolution layers with different convolution kernels; a model evaluation module, which is used to perform index evaluation on the multiple temperature prediction models based on the multiple validation sets to obtain multiple evaluation results; a model selection module, which is used to determine the target prediction model based on the multiple evaluation results.

[0013] In a fourth aspect, the present application further provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above method when executing the computer program.

[0014] In a fifth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, and the computer program implements the steps of the above method when executed by a processor.

[0015] The construction method, temperature prediction method, construction device, computer equipment and storage medium of the transformer winding temperature prediction model are used to train multiple convolutional neural network models respectively through multiple training sets determined by multiple data sets, thereby obtaining multiple temperature prediction models, and then the multiple temperature prediction models are evaluated for indicators according to multiple validation sets, and multiple evaluation results are obtained, and finally the target prediction model is determined based on the multiple evaluation results. Through this cross-validation method, the most suitable target prediction model can be determined, so as to have a more comprehensive understanding of the performance of the model on different data sets and reduce the deviation caused by data division. And by setting the convolutional neural network model to a structure including multiple parallel convolutional layers with different convolution kernels, in processing large-scale, high-dimensional temperature field data tasks, key features can be accurately refined, and the robustness of the model in feature extraction can be enhanced, thereby accelerating the learning process of the model, achieving a significant improvement in model performance, and more accurately predicting the temperature distribution of the transformer winding. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] FIG1 is an application environment diagram of a construction method and a temperature prediction method in one embodiment;

[0017] FIG2 is a flow chart of a method for constructing a transformer winding temperature prediction model in one embodiment;

[0018] FIG3 is a schematic diagram of a process for determining winding temperature sample data in one embodiment;

[0019] FIG4 is a schematic diagram of a process for obtaining a temperature prediction model in one embodiment;

[0020] FIG5 is a schematic diagram of a process for determining a predicted temperature distribution in one embodiment;

[0021] FIG6 is an architecture diagram of a convolutional neural network model in one embodiment;

[0022] FIG7 is a schematic diagram of a process for determining a predicted temperature distribution in another embodiment;

[0023] FIG8 is a schematic diagram of a process for updating a convolutional neural network model in one embodiment;

[0024] FIG9 is a schematic flow chart of a transformer winding temperature prediction method according to an embodiment;

[0025] FIG10 is a schematic diagram of current transformer status data in one embodiment;

[0026] FIG11 is a temperature cloud diagram of the current winding temperature distribution in one embodiment;

[0027] FIG12 is a module diagram of a device for constructing a transformer winding temperature prediction model in one embodiment;

[0028] FIG13 is a module diagram of a transformer winding temperature prediction device in one embodiment;

[0029] FIG. 14 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0031] The method for constructing a transformer winding temperature prediction model and the method for predicting transformer winding temperature provided in the embodiment of the present application can be applied in an application environment as shown in FIG1. ​​Among them, the terminal 102 communicates with the server 104 via a network. The data storage system can store data that the server 104 needs to process. The data storage system can be integrated on the server 104, or it can be placed on a cloud or other network server. The terminal 102 can be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, etc. The server 104 can be implemented with an independent server or a server cluster consisting of multiple servers.

[0032] In one embodiment, as shown in FIG. 2 , a method for constructing a transformer winding temperature prediction model is provided, and the method is applied to the terminal 102 in FIG. 1 as an example for explanation, and includes the following steps:

[0033] Step S210, obtaining a plurality of winding temperature sample data.

[0034] Specifically, the terminal 102 first obtains a plurality of winding temperature sample data for model training, and the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution. It can be understood that the inlet pressure and outlet pressure are the pressure of the transformer oil at the transformer oil inlet and outlet in the oil-immersed transformer, and the inlet temperature and outlet temperature are the temperature of the transformer oil at the transformer oil inlet and outlet in the oil-immersed transformer, wherein the transformer oil inlet and outlet can be one or more, which is determined by the specific structure of the transformer. The heat source parameter is the heat generation data of the transformer winding, and the windings on different sides can have different heat generation data. The winding temperature distribution is the temperature distribution on the winding under the corresponding conditions, which is the label in the winding temperature sample data. The winding temperature sample data can be collected by the sensor set on the transformer, or simulated by the transformer simulation model.

[0035] Step S220 , dividing the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets.

[0036] Specifically, after the terminal 102 obtains the plurality of winding temperature sample data, it first divides the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets. For example, the plurality of winding temperature sample data are divided into five non-overlapping subsets to obtain five data sets, and the number of winding temperature sample data in each data set is the same.

[0037] Step S230, determining multiple training sets and corresponding validation sets based on multiple data sets.

[0038] Specifically, after the terminal 102 divides the data set, it determines multiple training sets and corresponding validation sets based on the multiple data sets, and each data set is used as a validation set. That is, when the number of data sets is K, for each data set i, it is used as a validation set, and the remaining K-1 data sets are used as training sets. For example, when the number of data sets is 5, 5 training sets and corresponding validation sets can be determined.

[0039] Step S240: training multiple convolutional neural network models based on multiple training sets to obtain multiple temperature prediction models.

[0040] Specifically, after the terminal 102 determines multiple training sets and corresponding validation sets, it trains multiple convolutional neural network models according to the determined multiple training sets, thereby obtaining multiple temperature prediction models, wherein the convolutional neural network model includes multiple parallel convolutional layers with different convolutional kernels. For example, when it is determined that there are 5 training sets and corresponding validation sets, 5 convolutional neural network models are trained respectively by 5 training sets, thereby obtaining 5 trained temperature prediction models. It can be understood that the structures of multiple convolutional neural network models are the same. During training, a convolutional neural network model can also be iteratively trained independently by multiple training sets in sequence, thereby obtaining multiple trained temperature prediction models. Multiple parallel convolutional layers with different convolutional kernels in the convolutional neural network model are used to capture multi-level information in the input winding temperature sample data, and to more accurately extract key features.

[0041] Step S250, performing index evaluation on multiple temperature prediction models based on multiple validation sets to obtain multiple evaluation results.

[0042] Specifically, after obtaining multiple temperature prediction models, the terminal 102 performs index evaluation on the multiple temperature prediction models according to the validation set corresponding to the training set during training, thereby obtaining multiple corresponding evaluation results. The evaluation results may include: at least one of accuracy, precision, and recall rate.

[0043] Step S260, determining a target prediction model based on multiple evaluation results.

[0044] Specifically, after the terminal 102 obtains multiple evaluation results, it determines the best model according to the multiple evaluation results and uses it as the target prediction model. The target prediction model is the one that predicts the temperature distribution best among the multiple temperature prediction models.

[0045] The method for constructing the transformer winding temperature prediction model mentioned above trains multiple convolutional neural network models respectively through multiple training sets determined by multiple data sets, thereby obtaining multiple temperature prediction models, and then evaluates the indicators of the multiple temperature prediction models according to multiple validation sets, obtains multiple evaluation results, and finally determines the target prediction model based on the multiple evaluation results. Through this cross-validation method, the most suitable target prediction model can be determined, so as to have a more comprehensive understanding of the performance of the model on different data sets and reduce the deviation caused by data division. And by setting the convolutional neural network model to a structure including multiple parallel convolutional layers with different convolution kernels, in processing large-scale, high-dimensional temperature field data tasks, key features can be accurately refined, and the robustness of the model in feature extraction can be enhanced, thereby accelerating the learning process of the model, achieving a significant improvement in model performance, and more accurately predicting the temperature distribution of the transformer winding.

[0046] In one embodiment, as shown in FIG3 , the method for constructing a transformer winding temperature prediction model further includes:

[0047] Step S310, obtaining a plurality of original winding temperature data;

[0048] Step S320 , performing standardization processing on a plurality of original winding temperature data to obtain a plurality of winding temperature sample data.

[0049] Specifically, the winding temperature sample data in this embodiment is obtained by standardizing the original winding temperature data. Standardization is used to scale the data so that it falls into a small specific interval, usually [0,1] or [-1,1]. Standardization helps to speed up the convergence of the model and improve the training efficiency and prediction accuracy of the model. When performing standardization, first calculate the mean of each feature in the original winding temperature data. and standard deviation , and then each feature is standardized by the following formula:

[0050]

[0051] Among them, x is the original feature data, The standardized feature data is finally input into the convolutional neural network model for training. In some other embodiments, before the plurality of original winding temperature data are standardized, the original winding temperature data are also cleaned.

[0052] In one embodiment, as shown in FIG4 , in step S240, the step of training multiple convolutional neural network models based on multiple training sets respectively includes:

[0053] Step S241, input all winding temperature sample data in the training set into the convolutional neural network model to obtain the predicted temperature distribution.

[0054] Specifically, in this embodiment, in the process of training each convolutional neural network model, all winding temperature sample data in the training set are first input into the convolutional neural network model, and the predicted output of the model is calculated through forward propagation to obtain the predicted temperature distribution. The forward propagation includes calculation steps such as convolution layer, pooling layer, activation function, and fully connected layer.

[0055] Step S242, updating the model parameters of the convolutional neural network model based on the predicted temperature distribution and the winding temperature distribution to obtain a temperature prediction model.

[0056] Specifically, after obtaining the predicted temperature distribution, the winding temperature distribution in the winding temperature sample data is combined, and then the difference between the predicted temperature distribution and the actual winding temperature distribution is calculated according to the set loss function (such as mean square error MSE, mean absolute error MAE, etc.), and the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and then the model parameters of the convolutional neural network model are updated using an optimization algorithm (such as stochastic gradient descent SGD, Adam, etc.) to minimize the loss value, thereby obtaining a trained temperature prediction model.

[0057] In one embodiment, as shown in FIG5 , in step S241 , all winding temperature sample data in the training set are input into the convolutional neural network model to obtain the step of predicting the temperature distribution, including:

[0058] Step S243, performing convolution processing and activation processing on all winding temperature sample data to obtain a plurality of convolution data.

[0059] Specifically, after the winding temperature sample data is input into the convolutional neural network model, all the winding temperature sample data are first convolved and activated through multiple parallel convolutional layers with different convolutional kernels, thereby obtaining multiple convolutional data. As shown in FIG6 , it is an architecture diagram of a convolutional neural network model in an embodiment of the present application, which is provided with three parallel convolutional layers with different convolutional kernels, and the convolutional kernel sizes are 3*3, 5*5 and 7*7 respectively.

[0060] Step S244, performing pooling processing on the multiple convolution data to obtain multiple pooled data;

[0061] Specifically, after obtaining multiple convolution data, the multiple convolution data are pooled to obtain multiple pooled data. After downsampling through pooling, the amount of data and parameters can be compressed to reduce overfitting. For example, in FIG6 , the parallel convolution layers can independently perform convolution and activation processing on the input data, and the pooling layers can independently perform pooling processing to extract the pooled data at that scale, i.e., the feature map F i :

[0062]

[0063] in, is a complex function, I includes convolution processing, activation processing and pooling processing, K i is the parameter of the current convolution, the basic formula of the convolution operation in I can be expressed as:

[0064]

[0065] in, is the value of the convolution output, f is the input feature map, g is the convolution kernel, i and j are the offsets of the convolution kernel in the horizontal and vertical directions, m and n represent the width and height of the convolution kernel respectively, and b is the bias term, which can adjust the value of the convolution output. Represents an activation function. In some embodiments, when performing activation processing, a Swish activation function is used, and the specific formula is as follows:

[0066]

[0067] The Swish activation function is smooth and differentiable. During the back-propagation process, it can maintain the stable propagation of the gradient and reduce the risk of gradient vanishing or gradient exploding. This feature helps the model converge more stably during training, thereby improving the predictive performance of the regression model.

[0068] The pooling layer can use adaptive max pooling, which can convert inputs of any size into fixed-size outputs to ensure that all feature maps have the same size before merging. According to the specified output size, adaptive max pooling can dynamically divide the input feature map into regions of equal size and calculate the maximum value of each region to output the corresponding pooled data.

[0069] Let W and H be the width and height of the input feature map, respectively. W and O H is the desired width and height of the output feature map, k is the size of the pooling window, and s is the pooling step size. The goal of adaptive max pooling is to find a set of k and s values ​​so that the size of the output feature map is , to ensure that the size of the output feature map is consistent with expectations. Through the above processing, key features can be accurately extracted in processing large-scale, high-dimensional temperature field-level data tasks, the robustness of the model in feature extraction can be enhanced, and the learning process of the model can be accelerated, ultimately achieving a significant improvement in model performance.

[0070] Step S245, fusing the multiple pooled data to obtain fused data;

[0071] Specifically, through the above processing, multiple pooled data are obtained, that is, feature maps F i Finally, multiple pooled data are fused through the concatenation layer to obtain fused data. The fusion process uses simple feature concatenation, weighting and / or more complex nonlinear combination according to the specific situation. The fusion process can be expressed as:

[0072]

[0073] The Merge function depends on the selected fusion processing strategy. In some embodiments, a plurality of pooled data may be fused by feature concatenation to obtain fused data.

[0074] Step S246, determining the predicted temperature distribution based on the fused data.

[0075] Specifically, after obtaining the fused data, the final prediction regression task output is realized according to the fused feature map, thereby determining the predicted temperature distribution.

[0076] In one embodiment, as shown in FIG. 7 , in step S246 , the step of determining the predicted temperature distribution based on the fused data includes:

[0077] Step S247, performing a first full-connection process on the fused data to obtain first fully-connected data.

[0078] Specifically, as shown in Figure 6, after the fused data is obtained through the fusion process, the fused data is input into the first fully connected layer and the first fully connected process is performed to obtain the first fully connected data. The first fully connected process can perform more complex integration and transformation of the fused data, thereby improving the expression ability and performance of the model.

[0079] Step S248: Perform random data discarding processing on the first fully connected data to obtain second fully connected data.

[0080] Specifically, since the number of parameters of the fully connected layer is large, it is easy to produce overfitting. Therefore, after obtaining the first fully connected data, the first fully connected data is input into the random dropout layer for random data discarding, randomly discarding the outputs of some neurons, reducing the correlation between neurons, reducing the model's dependence on certain specific neurons, improving the generalization ability of the model, and reducing the risk of overfitting.

[0081] Step S249: performing a second full connection process on the second full connection data to obtain a predicted temperature distribution.

[0082] Specifically, after obtaining the second fully connected data, the second fully connected data is input into the second fully connected layer, and the second fully connected processing is performed, and the features output by the random inactivation layer are integrated into the final representation, so as to obtain the final predicted temperature distribution.

[0083] In one embodiment, as shown in FIG8 , in step S240, the step of training multiple convolutional neural network models based on multiple training sets respectively further includes:

[0084] Step S410, determining a learning rate by a cosine function;

[0085] Step S420, determining the speed and direction of updating the model parameters of the convolutional neural network model based on the learning rate.

[0086] Specifically, since temperature field-level data usually has the characteristics of high dimension and nonlinearity, in the training process of multi-scale convolutional neural network, the setting of learning rate directly affects the convergence speed and final performance of the model. Traditional fixed learning rate or linear attenuation strategy is difficult to adapt to the complex training process, which can easily lead to unstable training or fall into local optimal solution. Therefore, the embodiment of the present application dynamically adjusts the learning rate through the cosine function, so that the convolutional neural network model can converge quickly during the training process and effectively avoid local optimal solution. The specific formula for determining the learning rate through the cosine function is as follows:

[0087]

[0088] in, Represents the ratio of the learning rate after the tth iteration to the basic learning rate, and T represents the total number of iterations of model training. As a fine-grained learning rate control method, the cosine annealing learning rate decay strategy sets the learning rate to the initial value during training, and then gradually reduces the learning rate to the set minimum value through the cosine function waveform according to the preset period, and then increases it again to near the initial value, forming a periodic change until the end of training. This strategy cleverly balances the size of the learning rate, ensuring that the multi-scale convolutional neural network model can use a larger learning rate to quickly explore the parameter space in the early stage of training, reducing the time required for convergence; and in the middle and late stages of training, as the learning rate gradually decreases, the model can make more detailed adjustments near the optimal solution, improving the stability and final accuracy of the model.

[0089] In one embodiment, as shown in FIG9 , the present application embodiment further proposes a transformer winding temperature prediction method, including:

[0090] Step S510, obtaining current transformer status data.

[0091] Specifically, when predicting the transformer winding temperature, firstly, the current transformer state data of the transformer in the current state is obtained through the sensor. The current transformer state data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature and heat source parameters. It can be understood that the current transformer state data is the same type as the input data in the winding temperature sample data used in model training.

[0092] Step S520, inputting the current transformer state data into the target prediction model to obtain the current winding temperature distribution.

[0093] Specifically, after obtaining the current transformer state data, the current transformer state data is input into the target prediction model. After being processed by the target prediction model, the predicted temperature distribution of the winding, that is, the current winding temperature distribution, can be obtained. The target prediction model is obtained by the construction method of the transformer winding temperature prediction model in the above embodiment, and will not be described one by one here.

[0094] The transformer winding temperature prediction method of the present application is described in detail below with a specific embodiment. As shown in FIG10 , the transformer in the present embodiment is provided with one inlet and two outlets (the first outlet and the second outlet). In the current transformer state data collected, the inlet pressure is 0Pa, the inlet temperature is 300K, the outlet pressures of the first outlet and the second outlet are both 0Pa, and the outlet temperatures are both 300K. In the heat source parameters, the heat generation of the low voltage winding is 198400W / m 3 , the heat generated by the high voltage winding is 210000W / m 3. The heat generated by the loss of the oil-immersed transformer during operation causes the temperature of the windings, cores and other structural parts to increase continuously. Due to the large initial thermal conductivity of the two, the temperature increases rapidly. As the temperature of the transformer increases, the heat is transferred to the surface and the surrounding medium (mainly transformer oil) due to the temperature difference. Therefore, the transformer oil will get heat and the temperature will rise. At the same time, the temperature rise of the windings and cores will be controlled. There is a temperature difference inside the transformer. The oil density decreases due to the increase in temperature. The oil density in the oil tank tends to decrease from bottom to top. The low-density oil will move upward to form an oil flow cycle. When the oil flow flows, it will transfer heat to the wall of the transformer tank and the external radiator. After cooling, the oil flow will flow back to the bottom of the transformer tank through the oil channel. After the oil flow circulation process, the temperature of the transformer will not continue to rise. When it reaches a certain level, the heat generated inside the transformer is transferred to the surrounding environment through the oil flow, and its own temperature will no longer increase and tend to be stable. At this time, it will reach a thermal equilibrium state.

[0095] By inputting the collected current transformer state data into the target prediction model, the temperature distribution of the transformer winding under the current transformer state, that is, the current winding temperature distribution, is obtained. As shown in Figure 11, it is a temperature cloud map of the current winding temperature distribution predicted based on the current transformer state data in Figure 10. The target prediction model of the present application can realize the second-level calculation of the temperature field-level data based on the input current transformer state data, so as to achieve the real-time monitoring and accurate simulation of the transformer winding temperature, and the predicted temperature distribution is relatively accurate.

[0096] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0097] Based on the same inventive concept, the embodiment of the present application also provides a device for constructing a transformer winding temperature prediction model for implementing the method for constructing a transformer winding temperature prediction model involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in the embodiments of the device for constructing one or more transformer winding temperature prediction models provided below can be referred to the limitations of the method for constructing a transformer winding temperature prediction model above, and will not be repeated here.

[0098] In one embodiment, as shown in FIG. 12 , a device for constructing a transformer winding temperature prediction model is provided, comprising: a data acquisition module 610, a data division module 620, a data classification module 630, a model training module 640, a model evaluation module 650 and a model selection module 660, wherein:

[0099] The data acquisition module 610 is used to acquire a plurality of winding temperature sample data; wherein the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution;

[0100] A data partitioning module 620 is used to partition the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets;

[0101] A data classification module 630 is used to determine multiple training sets and corresponding validation sets based on multiple data sets, wherein each data set is used as a validation set;

[0102] A model training module 640 is used to train multiple convolutional neural network models based on multiple training sets to obtain multiple temperature prediction models; wherein the convolutional neural network model includes multiple convolutional layers with different convolution kernels;

[0103] A model evaluation module 650 is used to perform index evaluation on multiple temperature prediction models based on multiple validation sets to obtain multiple evaluation results;

[0104] The model selection module 660 is used to determine a target prediction model based on multiple evaluation results.

[0105] In one embodiment, the device for constructing a transformer winding temperature prediction model further includes:

[0106] The data processing module is used to obtain a plurality of original winding temperature data; perform standardization processing on the plurality of original winding temperature data to obtain a plurality of winding temperature sample data.

[0107] In one embodiment, the model training module 640 is also used to input all winding temperature sample data in the training set into the convolutional neural network model to obtain a predicted temperature distribution; based on the predicted temperature distribution and the winding temperature distribution, the model parameters of the convolutional neural network model are updated to obtain a temperature prediction model.

[0108] In one embodiment, the model training module 640 is also used to perform convolution processing and activation processing on all winding temperature sample data to obtain multiple convolution data; perform pooling processing on the multiple convolution data to obtain multiple pooling data; perform splicing processing on the multiple pooling data to obtain spliced ​​data; and determine the predicted temperature distribution based on the spliced ​​data.

[0109] In one embodiment, the model training module 640 is also used to perform a first full-connection processing on the spliced ​​data to obtain first fully-connected data; perform a random data discarding processing on the first fully-connected data to obtain second fully-connected data; and perform a second full-connection processing on the second fully-connected data to obtain a predicted temperature distribution.

[0110] In one embodiment, the model training module 640 is further used to determine a learning rate through a cosine function; and determine a speed and direction of updating model parameters of the convolutional neural network model based on the learning rate.

[0111] In one embodiment, as shown in FIG. 13 , a transformer winding temperature prediction device is provided, including: a data acquisition module 710 and a temperature prediction module 720, wherein:

[0112] The data acquisition module 710 is used to obtain the current transformer status data; wherein the current transformer status data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature and heat source parameters;

[0113] The temperature prediction module 720 is used to input the current transformer state data into the target prediction model to obtain the current winding temperature distribution; wherein the target prediction model is obtained by the construction method of the transformer winding temperature prediction model in the above embodiment.

[0114] The various modules in the above-mentioned transformer winding temperature prediction model construction device or transformer winding temperature prediction device can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0115] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be shown in FIG14. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, the memory, and the input / output interface are connected via a system bus, and the communication interface, the display unit, and the input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be implemented through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a method for constructing a transformer winding temperature prediction model or a transformer winding temperature prediction method is implemented. The display unit of the computer device is used to form a visually visible picture, which may be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0116] Those skilled in the art will understand that the structure shown in FIG. 14 is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0117] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0118] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the processor executes the computer program, the steps in the above-mentioned method embodiments are implemented.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for constructing a transformer winding temperature prediction model, characterized in that: The method comprises: Acquire a plurality of winding temperature sample data; wherein the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution; Dividing the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets; Determine multiple training sets and corresponding validation sets based on the multiple data sets; wherein each of the data sets is used as the validation set; Based on the multiple training sets, multiple convolutional neural network models are trained respectively to obtain multiple temperature prediction models; wherein the convolutional neural network model includes multiple parallel convolutional layers with different convolution kernels; Based on the multiple validation sets, the multiple temperature prediction models are evaluated for indicators to obtain multiple evaluation results; A target prediction model is determined based on the multiple evaluation results.

2. The method for constructing a transformer winding temperature prediction model according to claim 1, characterized in that: The method further comprises: Acquire multiple raw winding temperature data; The plurality of original winding temperature data are standardized to obtain a plurality of winding temperature sample data.

3. The method for constructing a transformer winding temperature prediction model according to claim 1, characterized in that: The step of respectively training multiple convolutional neural network models based on the multiple training sets includes: Inputting all the winding temperature sample data in the training set into the convolutional neural network model to obtain a predicted temperature distribution; The model parameters of the convolutional neural network model are updated based on the predicted temperature distribution and the winding temperature distribution to obtain the temperature prediction model.

4. The method for constructing a transformer winding temperature prediction model according to claim 3, characterized in that: The step of inputting all the winding temperature sample data in the training set into the convolutional neural network model to obtain a predicted temperature distribution includes: Performing convolution processing and activation processing on all the winding temperature sample data to obtain a plurality of convolution data; Performing pooling processing on the multiple convolution data to obtain multiple pooled data; Performing fusion processing on the plurality of pooled data to obtain fused data; The predicted temperature distribution is determined based on the fused data.

5. The method for constructing a transformer winding temperature prediction model according to claim 4, characterized in that: The step of determining the predicted temperature distribution based on the fused data comprises: Performing a first full-connection process on the fused data to obtain first full-connection data; Performing random data discard processing on the first fully-connected data to obtain second fully-connected data; A second full-connection process is performed on the second full-connection data to obtain the predicted temperature distribution.

6. The method for constructing a transformer winding temperature prediction model according to claim 3, characterized in that: The step of respectively training a plurality of convolutional neural network models based on the plurality of training sets further includes: Determine the learning rate through the cosine function; The speed and direction of updating the model parameters of the convolutional neural network model are determined based on the learning rate.

7. A method for predicting transformer winding temperature, characterized in that: The method comprises: Acquire current transformer status data; wherein the current transformer status data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature and heat source parameters; The current transformer state data is input into a target prediction model to obtain a current winding temperature distribution; wherein the target prediction model is obtained by the method for constructing a transformer winding temperature prediction model according to any one of claims 1 to 6.

8. A device for constructing a transformer winding temperature prediction model, characterized in that: The device comprises: A data acquisition module, used to acquire a plurality of winding temperature sample data; wherein the winding temperature sample data includes: inlet pressure, inlet temperature, outlet pressure, outlet temperature, heat source parameters and corresponding winding temperature distribution; A data partitioning module, used for partitioning the plurality of winding temperature sample data into a plurality of non-overlapping subsets to obtain a plurality of data sets; A data classification module, used to determine a plurality of training sets and corresponding validation sets based on the plurality of data sets; wherein each of the data sets is used as the validation set; A model training module, used to train multiple convolutional neural network models based on the multiple training sets to obtain multiple temperature prediction models; wherein the convolutional neural network model includes multiple convolutional layers with different convolution kernels; A model evaluation module, used to perform index evaluation on the plurality of temperature prediction models based on the plurality of validation sets to obtain a plurality of evaluation results; A model selection module is used to determine a target prediction model based on the multiple evaluation results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images