Transformer oil temperature prediction method and system

Abstract

The invention provides a transformer oil temperature prediction method and system, and belongs to the technical field of prediction, and the method comprises the steps: determining a target transformer monitoring data set; inputting the target transformer monitoring data set into the target tense fusion converter model to obtain a target transformer oil temperature prediction value; wherein the target temporal fusion converter model is obtained by training the temporal fusion converter model according to historical monitoring data of the plurality of transformers. According to the transformer oil temperature prediction method and system provided by the invention, the prediction precision of the transformer oil temperature can be improved, and the stability and reliability of the prediction result are ensured.

Description

Technical Field

[0001] The present disclosure belongs to the field of prediction technology, and more specifically, relates to a transformer oil temperature prediction method and system. Background Art

[0002] Transformers are core equipment in power systems, and their stable operation is crucial to the safety of the entire power grid. Transformer oil temperature is an important indicator of transformer health. Excessive oil temperature will affect the equipment's pressure resistance and mechanical strength, causing breakdown accidents. Therefore, accurate prediction of transformer oil temperature is of great significance to ensure the safe operation of power systems. Traditional oil temperature prediction methods mostly rely on empirical formulas and simple statistical models. These methods are difficult to cope with complex and changeable power system operating conditions, and the prediction accuracy is limited. Summary of the invention

[0003] The purpose of the present disclosure is to provide a transformer oil temperature prediction method and system to improve the prediction accuracy of the transformer oil temperature and ensure the stability and reliability of the prediction results.

[0004] A first aspect of an embodiment of the present disclosure provides a transformer oil temperature prediction method, comprising: Determine the target transformer monitoring dataset; Inputting the target transformer monitoring data set into the target temporal fusion converter model to obtain a target transformer oil temperature prediction value; The target temporal fusion converter model is obtained by training the temporal fusion converter model based on historical monitoring data of multiple transformers.

[0005] A second aspect of the embodiments of the present disclosure provides a transformer oil temperature prediction system, comprising: A data set determination module, used to determine a target transformer monitoring data set; An oil temperature prediction module, used for inputting the target transformer monitoring data set into a target temporal fusion converter model to obtain a target transformer oil temperature prediction value; The target temporal fusion converter model is obtained by training the temporal fusion converter model based on historical monitoring data of multiple transformers.

[0006] According to a third aspect of an embodiment of the present disclosure, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the steps of the above-mentioned method for predicting transformer oil temperature when executing the computer program.

[0007] According to a fourth aspect of the embodiments of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the above-mentioned transformer oil temperature prediction method are implemented.

[0008] The beneficial effects of a transformer oil temperature prediction method and system provided by the embodiments of the present disclosure are: The present disclosure uses a target temporal fusion transformer model (i.e., a target TFT model) for prediction. The model is trained based on historical monitoring data of multiple transformers and can fully capture the oil temperature change trends and laws of the transformer, thereby realizing accurate prediction of the target transformer oil temperature. It not only improves the prediction accuracy, but also helps to detect abnormal oil temperature in a timely manner. At the same time, by constructing a target TFT model, the present disclosure can not only capture short-term dynamic changes, but also understand long-term trends and periodic pattern changes, enhance the model's ability to represent input data, and further improve computational efficiency and model prediction accuracy. The present disclosure can effectively support on-site operation and maintenance in carrying out equipment status monitoring and preventive maintenance work through real-time prediction of oil temperature changes by the target TFT model. Therefore, the present disclosure can improve the prediction accuracy of transformer oil temperature and ensure the stability and reliability of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0010] Figure 1 A schematic diagram of a flow chart of a transformer oil temperature prediction method provided by an embodiment of the present disclosure; Figure 2 A schematic diagram of a TFT network model training process provided by an embodiment of the present disclosure; Figure 3 A flow chart of a transformer oil temperature prediction method based on multi-view feature extraction and TFT network modeling provided in an embodiment of the present disclosure Figure 4 A structural block diagram of a transformer oil temperature prediction system provided by an embodiment of the present disclosure; Figure 5 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0011] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0012] In order to make the purpose, technical solutions and advantages of the present disclosure more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0013] Please refer to Figure 1 , Figure 1 A schematic flow chart of a transformer oil temperature prediction method provided in an embodiment of the present disclosure, the method comprising: S101: Determine a target transformer monitoring data set.

[0014] In this embodiment, the target transformer is the transformer or a group of transformers for which oil temperature prediction is required. In actual power system and other application scenarios, there are often multiple transformers running at the same time, and the target transformer is a specific transformer individual (or a group of specific transformers) selected as the object of oil temperature prediction.

[0015] The monitoring data set is the data information obtained by real-time or periodic measurement and recording of the operating status of the target transformer. The above data set can reflect the various operating characteristics of the target transformer at different times, such as the input and output voltage, current size, operating environment temperature, load conditions and other aspects of the target transformer.

[0016] The target transformer monitoring data set includes operating parameters, meteorological parameters, operation and maintenance parameters, and derived parameters.

[0017] In this embodiment, determining the target transformer monitoring data set includes: Obtain monitoring data of the target transformer; Clean and standardize the monitoring data of the target transformer to obtain standard monitoring data of the target transformer; Performing data expansion on the standard monitoring data of the target transformer to obtain an expanded monitoring data set of the target transformer; The extended monitoring data set of the target transformer is subjected to feature screening to obtain the target transformer monitoring data set.

[0018] The process of determining the target transformer monitoring data set is the process of preprocessing the monitoring data of the target transformer. The monitoring data of the target transformer is obtained within a period of time and is used to establish the monitoring data set, and the target transformer monitoring data set is used as the input data set.

[0019] Data cleaning is used to process missing values ​​and outliers in monitoring data. Newton interpolation and Laida criterion can be used. Standardization is used to eliminate dimensional differences between different features. Z-score can be used for standardization. Data expansion can be analyzed according to the characteristics of the monitoring data. Dynamic feature parameters such as oil temperature, winding temperature, load rate, and load are derived using mathematical statistics methods for expansion. Derived features can include the temperature difference at the previous moment, the temperature difference at the previous two moments, the temperature difference at the same moment the day before, parameter temperature rise, season, cycle, daily maximum load rate, daily average load rate, load fluctuation rate, etc. Feature screening is used to select the features with the highest correlation between parameters. The maximum mutual information coefficient (MIC) can be used to calculate the correlation between parameters.

[0020] After a series of data preprocessing processes, the target transformer monitoring data set can be obtained.

[0021] S102: Input the target transformer monitoring data set into the target temporal fusion converter model to obtain the target transformer oil temperature prediction value; wherein the target temporal fusion converter model is obtained by training the temporal fusion converter model based on the historical monitoring data of multiple transformers.

[0022] In this embodiment, the target temporal fusion transformer (TFT) model is obtained by training the TFT model based on historical monitoring data. The historical monitoring data is the state data of multiple transformers in a period of time, which can reflect the various conditions of multiple transformers at different times and in different operating states in the past period of time. The TFT model is trained to learn the characteristics between the transformer monitoring data and the transformer oil temperature and the correlation between the data. This embodiment continuously adjusts the internal parameters of the TFT model and other mechanisms, so that the TFT model has the ability to generate reasonable outputs based on given inputs.

[0023] The target transformer monitoring data set is used as the input data set of the target TFT model, and the target transformer oil temperature prediction value is used as the output data set of the target TFT model. Therefore, the target transformer monitoring data set is input into the target TFT model to obtain the target transformer oil temperature prediction value. The oil temperature can be the temperature of the insulating oil inside the transformer.

[0024] It can be concluded from the above that the present disclosure uses a target temporal fusion transformer model (i.e., a target TFT model) for prediction. The model is trained based on historical monitoring data of multiple transformers and can fully capture the oil temperature change trends and laws of the transformer, thereby realizing accurate prediction of the target transformer oil temperature. It not only improves the prediction accuracy, but also helps to detect abnormal oil temperature in a timely manner. At the same time, by constructing a target TFT model, the present disclosure can not only capture short-term dynamic changes, but also understand long-term trends and periodic pattern changes, enhance the model's ability to represent input data, and further improve computational efficiency and model prediction accuracy. The present disclosure can effectively support on-site operation and maintenance to carry out equipment status monitoring and preventive maintenance work through real-time prediction of oil temperature changes by the target TFT model. Therefore, the present disclosure can improve the prediction accuracy of transformer oil temperature and ensure the stability and reliability of the prediction results.

[0025] In one embodiment of the present disclosure, a transformer oil temperature prediction method further includes: Identify historical monitoring data sets; Divide the historical monitoring data set into training set, validation set and test set; Perform feature extraction on the training set to obtain multi-view features; The temporal fusion transformer model is trained based on multi-view features to obtain a trained temporal fusion transformer model, and the trained temporal fusion transformer model is the target temporal fusion transformer model.

[0026] In this embodiment, the historical monitoring data set is data obtained after data preprocessing, including operating parameters, meteorological parameters, operation and maintenance parameters, and derived parameters.

[0027] The training set is a data set used to train the model. The model learns from the data in the training set to discover the various characteristics, rules, and correlations between the data, thereby continuously adjusting its own parameters and improving its processing capabilities for corresponding tasks (such as transformer oil temperature prediction).

[0028] The validation set is a data set used to verify the performance and effect of the model during the model training process. After the model has been trained for a certain period of time based on the training set, the validation set data will be used to input the model to check the degree of consistency between the model output results and the actual situation, thereby helping to determine whether the model has problems such as overfitting and assist in adjusting the model's training strategy, such as adjusting the learning rate and deciding whether to continue training.

[0029] The test set is used to evaluate the performance of the model and the accuracy of prediction after the model training is completed. Its data does not participate in the parameter adjustment of the model during the training process and is independent of the training set and the validation set. By inputting the test set data into the trained model and comparing the model output results with the actual data of the test set, the performance level of the model in the actual application scenario can be intuitively judged.

[0030] The historical monitoring data set is divided into training set, validation set and test set in a ratio of 7:2:1. The training set is mainly used for model training.

[0031] Feature extraction is the process of mining representative feature information that is helpful for transformer oil temperature prediction from the training set of historical monitoring data sets, which can be extracted using a neural network model.

[0032] Multi-view features are a collection of time series features, global features, and local features. They represent features extracted from multiple different angles and dimensions. They can more comprehensively reflect the operating status of the transformer and its relationship with the oil temperature, and help improve the effect of model training.

[0033] The data set of multi-view features is used as the input data set of the TFT model, and the transformer oil temperature prediction value is used as the output data set of the TFT model. The TFT model is trained to obtain a trained TFT model, namely, a target TFT model.

[0034] From the above, it can be concluded that this embodiment ensures the comprehensiveness and accuracy of model training by dividing the monitoring data set; then performs feature extraction to mine data value from multiple perspectives, thereby enhancing the generalization ability of the model; finally, the TFT model is trained based on multi-perspective features to obtain a highly optimized target model. Therefore, not only the accuracy of oil temperature prediction is improved, but also the stability and reliability of the model are enhanced, providing more accurate prediction support for the safe operation of the transformer, which has important practical application value.

[0035] In one embodiment of the present disclosure, determining a historical monitoring data set includes: Obtain historical monitoring data of multiple transformers within a first time period; Clean and standardize historical monitoring data to obtain historical standard monitoring data; Performing data expansion on historical standard monitoring data to obtain historical expanded monitoring data; The historical expanded monitoring data is subjected to feature screening to obtain the historical monitoring data set.

[0036] In this embodiment, the first time period is a period of time, which is dynamically selected according to the actual situation. The historical monitoring data of multiple transformers within a period of time are obtained, and the monitoring parameters are selected to establish a data set S. The main input parameters include winding temperature, load current, ambient temperature, wind speed, humidity, rainfall, equipment operating years, and defect records. The output parameter is oil temperature.

[0037] Data cleaning uses Newton interpolation method and Laida criterion to deal with missing values ​​and outliers in the data set respectively.

[0038] The Laida criterion judgment formula is:

[0039] in, represents the average value of the monitoring parameter in the data set S, represents the standard deviation, Indicates A monitoring parameter value.

[0040] The calculation formula of Newton interpolation method is:

[0041] in, This is the difference you want to calculate; are the coefficients of the difference polynomial, obtained by the difference quotient; These are all known parameter values.

[0042] This example uses Z-score to standardize the data set and eliminate the dimensional differences between different features. The calculation formula is:

[0043] in, represents the standardized result, represents the average value of the monitoring parameter in the data set S, represents the standard deviation, Indicates A monitoring parameter value.

[0044] According to the data set characteristic analysis, this embodiment uses mathematical statistics methods to derive a dynamic characteristic parameter expansion data set S for characteristic parameters such as oil temperature, winding temperature, load rate, and load. The derived characteristics mainly include the temperature difference at the previous moment, the temperature difference at the previous two moments, the temperature difference at the same moment the day before, parameter temperature rise, season, cycle, daily maximum load rate, daily average load rate, load fluctuation rate, etc.

[0045] This embodiment uses MIC to calculate the correlation between state parameters and selects the features with the highest correlation, that is, the features with the highest correlation coefficient MIC value. The state quantity is defined as x and y, and a certain number of intervals are divided in the x and y directions of the two-dimensional space. The calculation formula is as follows:

[0046] in, is the number of grids divided in the x and y directions, According to the actual training effect, it is most appropriate to take the 0.65th power of the number of data records. The filtered feature set (i.e., historical monitoring data set) is shown in Table 1 below. The historical monitoring data set includes operating parameters, meteorological parameters, operation and maintenance parameters, and derived parameters. Derived parameters are also derived features. Derived parameters are obtained using mathematical statistics methods based on operating parameters, meteorological parameters, and operation and maintenance parameters.

[0047] Table 1 Historical monitoring data set

[0048] It can be concluded from the above that in the process of determining the historical monitoring data set, this embodiment obtains the historical monitoring data of multiple transformers in the first time period, and undergoes a series of processes such as data cleaning, standardization, expansion and feature screening, and finally obtains a historical monitoring data set containing operating parameters, meteorological parameters, operation and maintenance parameters and derived parameters. This embodiment significantly improves the accuracy and availability of the data. At the same time, the data expansion and feature screening steps effectively enrich the content of the data set, which helps to more comprehensively reflect the actual operating status of the transformer and improve the accuracy of status monitoring and fault warning.

[0049] In one embodiment of the present disclosure, feature extraction is performed on the training set to obtain multi-view features, further comprising: Based on different sliding window lengths, the training set is intercepted to generate multiple data sequences of different scales; Extract data sequences of multiple different scales based on the time series model to obtain time series features; Based on the multi-layer perceptron, multiple data sequences of different scales are extracted to obtain global features; Based on the convolutional neural network model, multiple data sequences of different scales are extracted to obtain local features; Multi-view features include temporal features, global features and local features.

[0050] In one embodiment of the present disclosure, the time series model includes: adding a time attention weight in the input gate.

[0051] In this embodiment, the development process of the transformer oil temperature generally has a long-term repetitive pattern with certain regular changes related to seasons, day and night, etc., and a short-term repetitive pattern with oscillating changes due to power, weather, etc. The feature extraction of a single perspective cannot fully tap the implicit information in the feature parameters. Therefore, this embodiment uses a time series (Long Short-Term Memory, LSTM) model, a multi-layer perceptron (Multilayer Perceptron, MLP) and a convolutional neural network (Convolutional Neural Networks, CNN) model to extract multi-perspective feature extraction of the time series feature vector set, the global feature vector set and the local feature vector set in the training set of the historical monitoring data set. Different from the traditional series cross-time feature extraction, the parallel extraction of this embodiment can ensure the independence of the time series features, global features and local features in the original data, and can further improve the performance of the subsequent prediction model. The original data is a historical monitoring data set that has not undergone feature extraction.

[0052] Specifically, the steps of this embodiment are as follows: First, the divided training set is intercepted using different sliding window lengths to generate multiple data sequences of different scales. The sliding window length is randomly generated in three ways: short, medium, and long. The length of the short window is a random number in [6, 50]; the length of the medium window is a random number in [51, 500]; and the length of the long window is a random number in [501, 1000].

[0053] Secondly, the LSTM model is improved by adding the time attention weight to the input gate and adding the cell state at the previous moment to the forget gate to capture the temporal characteristics of the long-term dependency in the training data set. The calculation formula of the LSTM model is as follows:

[0054] in, , , They are input gate, forget gate and output gate respectively. is the cell state, , , , and are the corresponding weight matrices, and are the input value and output value at time t, respectively. is an additional weight matrix that allows the cell state to directly affect the forget gate, , , and are the corresponding bias terms respectively.

[0055] A dropout rate of 0.5 is introduced during training to control model complexity and avoid overfitting, while promoting rapid convergence of the model and improving training efficiency and model generalization ability.

[0056] The improved LSTM model is used to extract data sequences of multiple different scales to obtain time series features.

[0057] Then, a two-layer MLP model is constructed to obtain multi-level global feature information in the historical data set. The output calculation formula of the convolutional layer is as follows:

[0058] in, For the Tier The output value of a neuron, For the Layer Neuron to Layer The weight of the neuron, For the Tier The bias term of neurons, For the Tier The activation value of a neuron, To monitor the original values ​​of parameters (i.e., historical standard monitoring data), the propagation of original data helps the model better capture global characteristics.

[0059] A dropout rate of 0.5 is introduced during training to control model complexity and avoid overfitting, while promoting rapid convergence of the model and improving training efficiency and model generalization ability.

[0060] Finally, when this embodiment uses CNN for local feature extraction, the TimeDistributed layer is used to wrap the original convolution layer, and the sequence in the training data set is feature mapped point by point to increase the feature dimension of the original data and enrich the expression of feature information, ensuring that the model can focus on the features of each time step. Then, based on the two-dimensional convolutional neural network, the mapped feature map is deeply extracted through multi-layer convolution and pooling to obtain key local feature information. The pooling operation uses maximum pooling, and the calculation formula is as follows:

[0061] in, is the sliding step length, the value is 1, is the window size of the pooling operation, which is 2×2. and Relative offset within the window.

[0062] From the above, it can be concluded that this embodiment extracts multi-view features from the training set, uses sliding windows of different lengths to generate multi-scale data sequences, and uses time series models, multi-layer perceptrons, and convolutional neural network models for feature extraction, thereby obtaining rich temporal features, global features, and local features. This embodiment not only enhances the model's comprehensive understanding and analysis capabilities of data, but also improves the diversity and accuracy of feature representation. The comprehensive use of multi-view features helps the model better capture the potential laws and patterns in the data, thereby significantly improving the model's predictive performance and generalization capabilities.

[0063] In one embodiment of the present disclosure, a temporal fusion transformer model is trained based on multi-view features, including: Perform weighted calculation on temporal features, global features and local features to obtain fusion features; The temporal fusion transformer model is trained based on the fusion features to obtain an initial temporal fusion transformer model.

[0064] In this embodiment, in order to integrate different prediction step length requirements, a fusion feature is added to improve the accuracy of different prediction length models. The fusion formula is as follows:

[0065] in, To fusion features, , , are respectively the time series feature, global feature and local feature vector sets, , , When training the intraday prediction model, the initial weights can be set to 0.4, 0.1, 0.5; when training the intramonth prediction model, the initial weights can be set to 0.4, 0.2, 0.4; when training the long-term prediction model, the initial weights can be set to 0.4, 0.4, 0.2; during the training and tuning process, an adaptive adjustment strategy can be established based on the input data characteristics of different devices to follow the specific step size. The adjustment formula is as follows:

[0066] in, and is a parameter term, which is dynamically calculated by the gradient descent algorithm. is the step length.

[0067] Establishing a TFT model based on multi-feature extraction results can reduce the input dimension of the TFT model, reduce model complexity, and speed up training.

[0068] Specifically, see Figure 2 , based on the fusion features, the temporal fusion transformer model is trained to obtain the initial temporal fusion transformer model. The steps are as follows: First, the original TFT network model structure is improved by removing the original input data feature processing and variable selection network modules in the model, and directly using the encoder to receive fusion features of different scales.

[0069] Second, the attention mechanism in the original model is improved. The self-attention mechanism is modified into a multi-scale multi-head attention mechanism. For the fusion features of each scale, the multi-head attention mechanism is applied, the attention weight is calculated, and a weighted representation is generated. The formula of the multi-head attention mechanism is:

[0070] ; The calculation formula of the attention mechanism is: ; where Q, K, and V correspond to the query, key, and value vectors in the encoder or decoder, respectively. is the attention of each head, , , are the corresponding linear transformation matrices, is the dot product between the query and the key, is the dimension of the key vector.

[0071] Third, the decoder receives the output of the encoder and partial information of the target sequence, and uses a multi-scale multi-head attention mechanism to predict the value of the next time step. The output of the output layer decoder is fed into a fully connected layer, and then through an activation function and an output layer to obtain the final oil temperature prediction value.

[0072] From the above, it can be concluded that this embodiment forms fusion features by fusing time series features, global features and local features, and trains the temporal fusion transformer model based on this, which significantly improves the comprehensive analysis ability of the model. The fusion feature not only contains the key information of the data from different perspectives, but also balances the importance of each feature through weighted calculation, so that the model can understand the data more comprehensively. In addition, the training process based on the fusion feature enables the initial temporal fusion transformer model to have stronger learning ability and adaptability, laying a solid foundation for subsequent data processing and prediction tasks, and helping to achieve more efficient and accurate analysis and decision-making.

[0073] In one embodiment of the present disclosure, a transformer oil temperature prediction method further includes: The initial temporal fusion transformer is evaluated and adjusted based on the validation set to generate a trained temporal fusion transformer.

[0074] In this embodiment, the prediction ability of the model is evaluated using the validation set, and the difference between the predicted value and the true value is calculated using the minimization quantile as the optimization target. The optimization uses the gradient descent method to minimize the loss function, update the network weights, and obtain the final oil temperature prediction model. The quantile loss calculation formula is:

[0075] in, is the actual oil temperature value, is the predicted oil temperature value, is the quantile.

[0076] It can be concluded from the above that this embodiment can ensure the accuracy and reliability of the initial temporal fusion transformer in practical applications by strictly evaluating and adjusting the initial temporal fusion transformer on the validation set. This embodiment not only optimizes the parameter setting of the model, but also improves its generalization ability, so that the trained temporal fusion transformer can more accurately predict the transformer oil temperature, providing a strong guarantee for the safe and stable operation of the power system.

[0077] Figure 3 A flow chart of a transformer oil temperature prediction method based on multi-view feature extraction and TFT network modeling provided in an embodiment of the present disclosure includes: First, the dataset is constructed; Second, the sample set is divided; Third, multi-view feature extraction; Fourth, TFT network model training; Fifth, model optimization; Sixth, result prediction.

[0078] Corresponding to a transformer oil temperature prediction method in the above embodiment, Figure 4 This is a structural block diagram of a transformer oil temperature prediction system provided by an embodiment of the present disclosure. For ease of explanation, only the parts related to the embodiment of the present disclosure are shown. Figure 4 The transformer oil temperature prediction system 20 includes: a data set determination module 21 and an oil temperature prediction module 22.

[0079] Wherein, the data set determination module 21 is used to determine the target transformer monitoring data set; The oil temperature prediction module 22 is used to input the target transformer monitoring data set into the target temporal fusion converter model to obtain the target transformer oil temperature prediction value; Among them, the target temporal fusion converter model is obtained by training the temporal fusion converter model based on historical monitoring data of multiple transformers.

[0080] In one embodiment of the present disclosure, a transformer oil temperature prediction system 20 further includes: a model training module; A model training module, used to determine the historical monitoring data set; Divide the historical monitoring data set into training set, validation set and test set; Perform feature extraction on the training set to obtain multi-view features; The temporal fusion transformer model is trained based on multi-view features to obtain a trained temporal fusion transformer model, and the trained temporal fusion transformer model is the target temporal fusion transformer model.

[0081] In one embodiment of the present disclosure, the model training module is specifically used to: Obtain historical monitoring data of multiple transformers within a first time period; Clean and standardize historical monitoring data to obtain historical standard monitoring data; Performing data expansion on historical standard monitoring data to obtain historical expanded monitoring data; The historical expanded monitoring data is subjected to feature screening to obtain the historical monitoring data set.

[0082] In one embodiment of the present disclosure, the model training module is further used to: Based on different sliding window lengths, the training set is intercepted to generate multiple data sequences of different scales; Extract data sequences of multiple different scales based on the time series model to obtain time series features; Based on the multi-layer perceptron, multiple data sequences of different scales are extracted to obtain global features; Based on the convolutional neural network model, multiple data sequences of different scales are extracted to obtain local features; Multi-view features include temporal features, global features and local features.

[0083] In one embodiment of the present disclosure, the time series model includes: adding a time attention weight in the input gate.

[0084] In one embodiment of the present disclosure, the model training module is further used to: perform weighted calculation on the time series features, the global features and the local features to obtain fusion features; The temporal fusion transformer model is trained based on the fusion features to obtain an initial temporal fusion transformer model.

[0085] In one embodiment of the present disclosure, the model training module is further used to: The initial temporal fusion transformer is evaluated and adjusted based on the validation set to generate a trained temporal fusion transformer.

[0086] See also Figure 5 , Figure 5 A schematic block diagram of an electronic device provided by an embodiment of the present disclosure. Figure 5 The electronic device 300 in the embodiment shown may include: one or more processors 301, one or more input devices 302, one or more output devices 303 and one or more memories 304. The processors 301, input devices 302, output devices 303 and memories 304 communicate with each other via a communication bus 305. The memory 304 is used to store computer programs, which include program instructions. The processor 301 is used to execute the program instructions stored in the memory 304. The processor 301 is configured to call the program instructions to execute the functions of each module / unit in the above-mentioned system embodiments, such as Figure 4 The functions of modules 21 to 22 are shown.

[0087] It should be understood that in the embodiment of the present disclosure, the processor 301 may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0088] The input device 302 may include a touch panel, a fingerprint collection sensor (for collecting the user's fingerprint information and fingerprint direction information), a microphone, etc., and the output device 303 may include a display (LCD, etc.), a speaker, etc.

[0089] The memory 304 may include a read-only memory and a random access memory, and provide instructions and data to the processor 301. A portion of the memory 304 may also include a non-volatile random access memory. For example, the memory 304 may also store information about the device type.

[0090] In a specific implementation, the processor 301, input device 302, and output device 303 described in the embodiments of the present disclosure can execute the implementation methods described in the first and second embodiments of a transformer oil temperature prediction method provided in the embodiments of the present disclosure, and can also execute the implementation methods of the electronic device described in the embodiments of the present disclosure, which will not be repeated here.

[0091] In another embodiment of the present disclosure, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by the processor, all or part of the processes in the above-mentioned embodiment method are implemented, and the computer program can also be completed by instructing the relevant hardware through the computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, the steps of each of the above-mentioned method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0092] The computer-readable storage medium may be an internal storage unit of the electronic device of any of the aforementioned embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium may also be an external storage device of the electronic device, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Furthermore, the computer-readable storage medium may also include both an internal storage unit of the electronic device and an external storage device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium may also be used to temporarily store data that has been output or is to be output.

[0093] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0094] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the electronic devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0095] In the several embodiments provided in the present application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces or units, or it can be an electrical, mechanical or other form of connection.

[0096] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of the present disclosure.

[0097] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0098] The above are only specific embodiments of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present disclosure, and these modifications or replacements should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.

Claims

1. A transformer oil temperature prediction method, characterized in that: include: Determine the target transformer monitoring dataset; Inputting the target transformer monitoring data set into the target temporal fusion converter model to obtain a target transformer oil temperature prediction value; The target temporal fusion converter model is obtained by training the temporal fusion converter model based on historical monitoring data of multiple transformers.

2. A transformer oil temperature prediction method according to claim 1, characterized in that: Also includes: Identify historical monitoring data sets; Dividing the historical monitoring data set into a training set, a validation set, and a test set; Performing feature extraction on the training set to obtain multi-view features; A temporal fusion transformer model is trained based on the multi-view features to obtain a trained temporal fusion transformer model, and the trained temporal fusion transformer model is the target temporal fusion transformer model.

3. A transformer oil temperature prediction method as claimed in claim 2, characterized in that: The determining of the historical monitoring data set includes: Obtain historical monitoring data of multiple transformers within a first time period; Performing data cleaning and standardization on the historical monitoring data to obtain historical standard monitoring data; Performing data expansion on the historical standard monitoring data to obtain historical expanded monitoring data; The historical expanded monitoring data is subjected to feature screening to obtain a historical monitoring data set.

4. A transformer oil temperature prediction method as claimed in claim 3, characterized in that: The extracting features from the training set to obtain multi-view features also includes: Based on different sliding window lengths, the training set is intercepted to generate multiple data sequences of different scales; Extracting the data sequences of the multiple different scales based on a time series model to obtain time series features; Extracting the data sequences of the multiple different scales based on a multi-layer perceptron to obtain global features; Extracting the data sequences of the multiple different scales based on a convolutional neural network model to obtain local features; The multi-view features include temporal features, global features and local features.

5. A transformer oil temperature prediction method as claimed in claim 4, characterized in that: The time series model includes: adding a time attention weight in an input gate.

6. A transformer oil temperature prediction method as claimed in claim 4, characterized in that: The training of the temporal fusion transformer model based on the multi-view features comprises: Performing weighted calculation on the temporal feature, the global feature and the local feature to obtain a fusion feature; The temporal fusion transformer model is trained based on the fusion features to obtain an initial temporal fusion transformer model.

7. A transformer oil temperature prediction method according to claim 6, characterized in that: Also includes: The initial temporal fusion transformer is evaluated and adjusted based on the validation set to generate a trained temporal fusion transformer.

8. A transformer oil temperature prediction system, characterized in that: include: A data set determination module, used to determine a target transformer monitoring data set; An oil temperature prediction module, used for inputting the target transformer monitoring data set into a target temporal fusion converter model to obtain a target transformer oil temperature prediction value; The target temporal fusion converter model is obtained by training the temporal fusion converter model based on historical monitoring data of multiple transformers.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: 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 storing a computer program, 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