Transformer top oil temperature prediction method based on comprehensive thermal factor and neural network

Abstract

The invention relates to a transformer top oil temperature prediction method based on a comprehensive thermal factor and a neural network, and the method comprises the steps: 1, selecting characteristic variables which affect the top oil temperature of a transformer through the actual measurement data of the transformer, and carrying out the correlation analysis, and obtaining a correlation heat map; 2, determining a characteristic number closely related to the top oil temperature thermal effect by combining a top oil temperature calculation formula in IEEE and GB1094, defining the characteristic number as a comprehensive thermal factor, and taking the comprehensive thermal factor and the top oil temperature as characteristic variables; 3, constructing a top oil temperature prediction model of a CNN-LSTM-Attention combined structure, constructing a linear connection layer as a comprehensive thermal factor by using the current and the environment temperature, and dividing a training set sample proportion to a verification set sample proportion to be 6: 2: 2; 4, the average absolute error and the average absolute percentage error serve as error evaluation indexes, and the top oil temperature prediction result is evaluated and analyzed. The method can accurately predict the top oil temperature of the transformer in a future period of time.

Description

Technical Field

[0001] The invention relates to the technical field of oil-immersed power transformers, in particular to a transformer top layer oil temperature prediction method based on comprehensive thermal factors and neural networks. Background Art

[0002] The transformer is responsible for the voltage change and power conversion between power grids. It is a key equipment in the power system and plays an important role in maintaining the normal operation of the entire system. Once the transformer fails, it may cause a power outage or even the loss of equipment assets, affecting the normal production and life. The insulation state of the transformer determines whether its operating state is good. The damage to the insulation state often leads to the occurrence of failures. The hot spot temperature of the transformer will directly affect its insulation aging speed, and the hot spot temperature is difficult to obtain. The correlation between the top oil temperature and the hot spot temperature of the winding is strong. If the top oil temperature of the transformer is too high, it will damage the insulation performance and greatly increase its operation uncertainty. Based on the historical monitoring data of the transformer, the thermal state of the top oil temperature is predicted and monitored, which can evaluate the thermal state of the transformer in advance and provide a certain reference for the judgment of its insulation state.

[0003] At present, the research on the prediction of the top oil temperature of power transformers mainly focuses on inputting a variety of different characteristic parameters into the prediction model after being screened by order reduction or correlation analysis. The physical meaning of the input model characteristic quantity is unclear. In terms of the prediction time scale, most studies focus more on the prediction results of the ultra-short time scale, and there are relatively few prediction studies conducted on multiple time steps. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method for predicting the top oil temperature of a transformer based on comprehensive thermal factors and neural networks, which can accurately predict the top oil temperature of the transformer in the future.

[0005] To achieve the above object, the present invention adopts the following technical solution: a transformer top oil temperature prediction method based on comprehensive thermal factors and neural network, comprising the following steps:

[0006] Step 1: Obtain transformer monitoring data, including characteristic quantities such as ambient temperature, load current, top oil temperature and active power, and use the obtained transformer monitoring data to perform correlation analysis on factors affecting the top oil temperature to obtain a correlation heat map of the top oil temperature;

[0007] Step 2: Based on the results of the thermal map correlation analysis, combined with IEEE.StdC57.91-1995 "IEEE Guide for Loading Mineral Oil-Immersed Transformers and Step-Voltage Regulators" and GB / T1094.7-2008 "Power Transformers Part 7: Guidelines for Loading Oil-Immersed Power Transformers" and the derived formula related to the calculation of the top oil temperature in GB1094, the characteristic number closely related to the thermal effect of the top oil temperature is determined, which is defined as the comprehensive thermal factor. It is proposed to use the top oil temperature and the comprehensive thermal factor as the input feature quantity for the prediction of the top oil temperature;

[0008] Step 3: Build a top oil temperature prediction model with a CNN-LSTM-Attention combination structure. After normalizing the proposed input feature parameters, construct a linear connection layer for the current and the ambient temperature, and divide the training set, validation set, and test set into a ratio of 6:2:2 to predict the top oil temperature for a period of time in the future.

[0009] Step 4: Use the mean absolute error and mean absolute percentage error as error evaluation indicators to evaluate the prediction effect of the top oil temperature of the comprehensive thermal factor and CNN-LSTM-Attention model.

[0010] In a preferred embodiment, in step 1, the historical monitoring data of the top layer oil temperature includes ambient temperature, load current, high voltage side active power and top layer oil temperature monitoring data, and the sampling interval of the samples is 0.5 hours.

[0011] In a preferred embodiment, in step 1, the correlation analysis method is the Pearson correlation coefficient method, and the correlation analysis of each feature quantity is performed by the Pearson correlation coefficient method. The calculation method of the Pearson correlation coefficient γ is:

[0012]

[0013] in, and represents the average value of the corresponding feature parameter sample, X j and Y j are the j-th values ​​of the feature samples, and n represents the total number of values.

[0014] In a preferred embodiment, in step 2, the top oil temperature calculation formula derived from the guideline is discretized, and after parameter equivalence using coefficients, the calculation physical model of the top oil temperature value is obtained as follows:

[0015]

[0016] Among them, θ top,k-1 is the top oil temperature value at the historical moment, θ amb,k is the ambient temperature, I k represents the load current, represents the square of the load current, θ top,k is the top oil temperature value at the predicted time, K 1 , K 2 , K 3 and K 4 are the coefficients to be determined respectively.

[0017] In a preferred embodiment, in step 2, the comprehensive thermal factor is the coupling of the square of the current and the ambient temperature. Considering that the oil flow circulation process of the transformer takes a certain amount of time, there is a certain time difference between the current rise and the top oil temperature rise; therefore, in the selection of the comprehensive thermal factor, the hysteresis effect of the current is also considered, and the calculation formula of the comprehensive thermal factor is:

[0018] Q=k 1 I k 2 +k 2 θ amb,k

[0019] Among them, I k represents the load current, θ amb,k represents the ambient temperature, Q is the defined comprehensive thermal factor, k 1 and k 2 are the coefficient parameters to be determined respectively.

[0020] In a preferred embodiment, in step 3, the comprehensive thermal factor and the historical monitoring data of the top oil temperature are normalized and input into the CNN-LSTM-Attention model. The two input quantities of the comprehensive thermal factor are connected through a linear connection layer, and k is adjusted in real time through the prediction process. 1 and k 2 The weight of and the top oil temperature historical monitoring data together constitute the input layer; the normalization processing method is:

[0021]

[0022] in, is the normalized variable value, x i is the original value of the variable, x max and x min are the maximum and minimum values ​​of the characteristic variable respectively.

[0023] In a preferred embodiment, in step 3, the neural network algorithm is a CNN-LSTM-Attention combination algorithm, the convolutional layer CNN can obtain effective information from the local area, the LSTM layer can effectively handle the dependency problems in long sequences, and the Attention layer can better understand the important correlations within the sequence.

[0024] In a preferred embodiment, in step 4, the error evaluation index is mean absolute error MAE, root mean square error RMSE and mean absolute percentage error MAPE, and the calculation method of the two evaluation indexes is:

[0025]

[0026] Among them, y i is the actual value of the top oil temperature, is the predicted value of the top oil temperature, and m represents the sample size for calculation.

[0027] In a preferred embodiment, in step 4, the modules of the combined model are connected using a fully connected layer, and the prediction results of the comprehensive thermal factors are evaluated using an error evaluation index to obtain a MAPE value of the prediction result of <5% and a MAE value of <1°C.

[0028] Compared with the prior art, the present invention has the following beneficial effects: On the basis of correlation analysis, the present invention considers the influence of the current hysteresis effect and incorporates it into the correlation analysis, which fully reflects that there is a certain time difference between the current rise and the top oil temperature rise during the transformer oil flow circulation process. The comprehensive thermal factor proposed in the present invention clarifies the physical meaning of the input feature quantity, and adjusts the coefficient weights through real-time feedback of the linear connection layer, avoiding the one-sidedness of fixed weight calculation. The CNN-LSTM-Attention combined model mechanism adopted by the present invention combines the advantages of three different models, and at the same time optimizes the hyperparameters of LSTM, which can effectively realize the prediction of the top oil temperature in the future. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a correlation coefficient heat map of an embodiment of the present invention.

[0030] Figure 2 It is a flow chart of the method implementation of the embodiment of the present invention.

[0031] Figure 3 This is a diagram showing the prediction results of the top oil temperature in the next 12 hours in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0033] It should be noted that the following detailed descriptions are illustrative and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present application belongs.

[0034] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application; as used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or their combinations.

[0035] See also Figure 1-3 The present invention provides a transformer top oil temperature prediction method based on comprehensive thermal factors and neural network, comprising the following steps:

[0036] Step 1: Using the acquired transformer monitoring data, including characteristic quantities such as ambient temperature, load current, top oil temperature and active power, a correlation analysis of factors affecting the top oil temperature is performed to obtain a correlation heat map of the top oil temperature.

[0037] The historical monitoring data of the top oil temperature include ambient temperature, load current, high voltage side active power and top oil temperature monitoring data, and the sampling interval of the samples is 0.5 hours. The correlation analysis method is the Pearson correlation coefficient method, and the correlation analysis of each feature quantity is performed by the Pearson correlation coefficient method. The calculation method of the Pearson correlation coefficient is:

[0038]

[0039] in, and represents the average value of the corresponding feature parameter sample, X j and Y j are the j-th values ​​of the feature samples respectively.

[0040] Step 2: Based on the results of the thermal map correlation analysis, combined with IEEE.StdC57.91-1995 "IEEE Guide for Loading Mineral Oil-Immersed Transformers and Step-Voltage Regulators" and GB / T1094.7-2008 "Power Transformers Part 7: Guidelines for Loading Oil-Immersed Power Transformers" and the derived formula related to the calculation of the top oil temperature in GB1094, the characteristic number closely related to the thermal effect of the top oil temperature is determined, which is defined as the comprehensive thermal factor. It is proposed that the top oil temperature and the comprehensive thermal factor be used as the input feature quantity for the prediction of the top oil temperature.

[0041] Considering that the oil circulation process of the transformer takes a certain amount of time, there is a certain time difference between the current rise and the top oil temperature rise. In order to further improve the accuracy of the top oil temperature prediction, the top oil temperature and the current with different lag times are analyzed with a time step of 1h when performing the correlation analysis. The top oil temperature calculation formula derived from the guideline is discretized, and the coefficients are used for parameter equivalence to obtain the physical model for the top oil temperature calculation as follows:

[0042]

[0043] Among them, θ top,k-1 is the top oil temperature value at the historical moment, θ amb,k is the ambient temperature, I k represents the load current, θ top,k is the top oil temperature value at the predicted time, K 1 , K 2 , K 3 and K 4 are the coefficients to be determined respectively.

[0044] The comprehensive thermal factor is the coupling of the square of the current and the ambient temperature. Combined with the current hysteresis effect mentioned above, the thermal factors with hysteresis times of 1h, 2h and 3h are proposed. The calculation formula of the comprehensive thermal factor is:

[0045] Q = k 1 I 2 +k 2 θ amb

[0046] Among them, I k represents the load current, θ amb,k represents the ambient temperature, Q is the defined comprehensive thermal factor, k 1 and k 2 are the coefficient parameters to be determined respectively.

[0047] Step 3: Construct a top oil temperature prediction model with a CNN-LSTM-Attention combined structure. After normalizing the proposed input feature parameters, construct a linear connection layer for the current and ambient temperature, and divide the training set, validation set, and test set into a ratio of 6:2:2 to predict the top oil temperature in the future.

[0048] In the CNN convolution calculation, the number of convolution kernels is set to 3 and the convolution size is set to 16. When optimizing the hyperparameters of the LSTM layer, grid search methods, genetic algorithms, etc. can be used. In this example, a simple Bayesian optimization method is used, with the minimum top oil temperature prediction loss value as the objective function, to obtain the optimized transformer top oil temperature prediction model.

[0049] Step 4: Use the mean absolute error and mean absolute percentage error as error evaluation indicators to evaluate the prediction effect of the top oil temperature of the comprehensive thermal factor and CNN-LSTM-Attention model.

[0050] The error evaluation indicators selected are mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). The calculation methods of the two evaluation indicators are:

[0051]

[0052] Among them, y i is the actual value of the top oil temperature, is the predicted value of the top oil temperature, and m represents the sample size for calculation.

[0053] The fully connected layer is used to connect the various modules of the combined model, and the error evaluation index is used to evaluate the prediction results of the comprehensive thermal factor.

[0054] The beneficial effects of the present invention are further illustrated below by taking a 220kV oil-immersed power transformer as an example.

[0055] 1) Obtain the detection data of the 220kV transformer from July to October 2023. Considering that the transformer oil circulation process takes a certain amount of time, with a step length of 1h, discuss the correlation between the monitoring data and the top oil temperature, such as Figure 1 As shown in the figure. When the current lag time is 1h, 2h and 3h, the correlation with the top oil temperature is the largest. In addition, the ambient temperature and the oil temperature at the historical moment are also used as characteristic variables. Since power is a function of current, it is not considered as an input variable.

[0056] 2) Combined with the relevant formulas for calculating the top oil temperature in the guidelines, a coupling method for calculating the comprehensive thermal factor is proposed. After the current and the ambient temperature are connected through a linear layer for weighted calculation, they are used together with the oil temperature at the historical moment as input feature quantities. Among them, the current in the comprehensive thermal factor considers the current values ​​under three different lag times.

[0057] 3) Construct a CNN-LSTM-Attention combined prediction model. The prediction process is as follows: Figure 2 As shown, the comprehensive thermal factor input conditions of the above three different lag times are predicted, and the hyperparameters in the combined model are optimized.

[0058] 4) The prediction results are as follows Figure 3 As shown in (a), MAE, RMSE and MAPE are used as error evaluation indicators to analyze the prediction results, as shown in Figure 3 (b) is shown in Table 1.

[0059] Table 1 Prediction error

[0060]

[0061] The above are preferred embodiments of the present invention. Any changes made according to the technical solution of the present invention, as long as the resulting functions do not exceed the scope of the technical solution of the present invention, belong to the protection scope of the present invention.

Claims

1. A transformer top oil temperature prediction method based on comprehensive thermal factors and neural networks, characterized in that: The following steps are involved: Step 1: Obtain transformer monitoring data, including characteristic quantities such as ambient temperature, load current, top oil temperature and active power, and use the obtained transformer monitoring data to perform correlation analysis on factors affecting the top oil temperature to obtain a correlation heat map of the top oil temperature; Step 2: Based on the results of the thermal map correlation analysis and combined with the derivation formula related to the calculation of top oil temperature in IEEE, the characteristic number closely related to the thermal effect of top oil temperature is determined, which is defined as the comprehensive thermal factor. It is proposed to use the top oil temperature and the comprehensive thermal factor as the input characteristic quantity for top oil temperature prediction; Step 3: Build a top oil temperature prediction model with a CNN-LSTM-Attention combination structure. After normalizing the proposed input feature parameters, construct a linear connection layer for the current and the ambient temperature, and divide the training set, validation set, and test set into a ratio of 6:2:2 to predict the top oil temperature for a period of time in the future. Step 4: Use the mean absolute error and mean absolute percentage error as error evaluation indicators to evaluate the prediction effect of the top oil temperature of the comprehensive thermal factor and CNN-LSTM-Attention model.

2. The transformer top oil temperature prediction method based on comprehensive thermal factor and neural network according to claim 1 is characterized in that: In step 1, the historical monitoring data of the top oil temperature includes ambient temperature, load current, high-voltage side active power and top oil temperature monitoring data, and the sampling interval of the samples is 0.5 hours.

3. The transformer top oil temperature prediction method based on comprehensive thermal factor and neural network according to claim 1 is characterized in that: In step 1, the correlation analysis method is the Pearson correlation coefficient method, and the correlation analysis of each feature quantity is performed by the Pearson correlation coefficient method. The calculation method of the Pearson correlation coefficient γ is: in, and represents the average value of the corresponding feature parameter sample, X j and Y j are the j-th values ​​of the feature samples, and n represents the total number of samples.

4. The transformer top oil temperature prediction method based on comprehensive thermal factor and neural network according to claim 1 is characterized in that: In step 2, the top oil temperature calculation formula derived from the guideline is discretized, and after using coefficients for parameter equivalence, the calculation physical model of the top oil temperature value is obtained as follows: Among them, θ top,k-1 is the top oil temperature value at the historical moment, θ amb,k is the ambient temperature, I k represents the load current, represents the square of the load current, θ top,k is the top oil temperature value at the predicted moment, and K1, K2, K3 and K4 are coefficients to be determined.

5. The transformer top oil temperature prediction method based on comprehensive thermal factor and neural network according to claim 1 is characterized in that: In step 2, the comprehensive thermal factor is the coupling of the square of the current and the ambient temperature. Considering that the oil flow circulation process of the transformer takes a certain amount of time, there is a certain time difference between the current rise and the top oil temperature rise; therefore, in the selection of the comprehensive thermal factor, the hysteresis effect of the current is also considered. The calculation formula of the comprehensive thermal factor is: Q=k1I k 2 +k2θ amb,k Among them, I k represents the load current, θ amb,k represents the ambient temperature, Q is the defined comprehensive thermal factor, and k1 and k2 are coefficient parameters to be determined.

6. The transformer top oil temperature prediction method based on comprehensive thermal factor and neural network according to claim 1 is characterized in that: In step 3, the comprehensive thermal factor and the historical monitoring data of the top oil temperature are normalized and input into the CNN-LSTM-Attention model. The two input quantities of the comprehensive thermal factor are connected through a linear connection layer, and the weights of k1 and k2 are adjusted through real-time feedback of the prediction process, and together with the historical monitoring data of the top oil temperature, they constitute the input layer; the normalization processing method is: in, is the normalized variable value, x i is the original value of the variable, x max and x min are the maximum and minimum values ​​of the characteristic variable respectively.

7. The transformer top oil temperature prediction method based on comprehensive thermal factors and neural network according to claim 1 is characterized in that: In step 3, the neural network algorithm is a CNN-LSTM-Attention combination algorithm, the convolutional layer CNN can obtain effective information from the local area, the LSTM layer can effectively handle the dependency problem in the long sequence, and the Attention layer can better understand the important correlation within the sequence.

8. The transformer top oil temperature prediction method based on comprehensive thermal factors and neural network according to claim 1 is characterized in that: In step 4, the error evaluation indicators are mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE). The calculation methods of the two evaluation indicators are: Among them, y i is the actual value of the top oil temperature, is the predicted value of the top oil temperature, and m represents the sample size for calculation.

9. The transformer top oil temperature prediction method based on comprehensive thermal factors and neural network according to claim 1 is characterized in that: In step 4, the modules of the combined model are connected using a fully connected layer, and the prediction results of the comprehensive thermal factors are evaluated using an error evaluation index, and the MAPE value of the prediction results is obtained to be <5%, and the MAE value is <1°C.

Images