Transformer oil temperature prediction method and device based on actual operation data

Abstract

The invention discloses a transformer oil temperature prediction method and device based on actual operation data, and relates to the technical field of electric power system operation and inspection. The method comprises the steps of obtaining actual operation data of a transformer; dividing the actual operation data into different transformer types according to the operation voltage grade and the cooling mode of the transformer; the actual operation data are preprocessed, missing values are filled, and abnormal values are eliminated; constructing an integrated learning model, and training the integrated learning model through the actual operation data of the same transformer type to obtain an oil temperature prediction model of each transformer type; and predicting the oil temperature of each type of transformer through the oil temperature prediction model of each type of transformer. The integrated learning model is used as a prediction model, different models are used for capturing different features in data, and the generalization ability and prediction precision of the models are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system operation and inspection, and in particular to a transformer oil temperature prediction method based on actual operation data and a device thereof. Background Art

[0002] The transformer is one of the important equipment in the power system, and its operating status is related to the safety and stability of the entire power system. Transformers can be divided into oil-immersed transformers and dry-cooled transformers according to the cooling method. Oil-immersed transformers use insulating oil as the cooling medium. Because of their superior insulation performance and good cooling effect, they are widely used in power systems. Dry-cooled transformers use air as the cooling medium and are usually used indoors and in flammable and explosive places. Transformer internal failures and reduced lifespan are usually caused by reduced insulation capacity, and the internal temperature change of the transformer is the main reason for the reduced insulation capacity. Therefore, accurate prediction of the internal temperature of the transformer is very important for the safety of the power system.

[0003] A large number of studies have used artificial intelligence algorithms to model and predict transformer oil temperature. Commonly used algorithms include BP neural network, linear regression model, support vector machine, etc. The influencing parameters of transformer oil temperature are selected through principal component analysis or association analysis. However, a single prediction model is prone to fall into a local optimal solution, so it is necessary to optimize hyperparameters. The process of optimizing hyperparameters consumes a lot of time and computing resources, and the model generalization ability is poor.

[0004] In terms of data sets, key data related to oil temperature of transformers, such as load, ambient temperature, internal oil temperature, etc., are mostly artificially idealized settings, with strong theoretical guidance significance but little practical engineering application value. The data sets targeted by current research are usually small-scale ideal data sets, with training sets usually in the hundreds of groups, and test sets and validation sets usually only in the dozens. The applicability of the prediction models established based on these data is poor and the generalization performance is weak. However, in actual engineering, the data scale of transformers is collected in minutes, the data scale is large, and the actual operating voltage, model, and cooling method of the transformer are also diverse. It is impossible to determine the prediction ability of these research methods on large-scale real data sets.

[0005] Based on the above problems, it is urgent to propose an optimized transformer oil temperature prediction method to improve the prediction accuracy. Summary of the invention

[0006] The present invention provides a transformer oil temperature prediction method based on actual operation data and a device thereof, so as to solve the problems of the existing prediction methods.

[0007] The present invention is achieved through the following technical solutions:

[0008] A first aspect of the present invention provides a transformer oil temperature prediction method based on actual operation data, comprising:

[0009] Acquire actual operation data of the transformer, wherein the actual operation data includes oil temperature and oil temperature influencing parameters;

[0010] Classifying the actual operating data into different transformer types according to the operating voltage level and cooling method of the transformer;

[0011] Preprocessing the actual operation data to fill missing values ​​and remove outliers;

[0012] Constructing an integrated learning model, training the integrated learning model through actual operation data of the same transformer type, and obtaining an oil temperature prediction model for each transformer type, wherein the input of the integrated learning model is an oil temperature influencing parameter, and the output is the oil temperature;

[0013] The oil temperature of each type of transformer is predicted by the oil temperature prediction model of each type of transformer.

[0014] The present invention aims at the problem that the training data of the existing prediction model is usually idealized and the actual prediction effect is difficult to verify. The prediction model is trained through actual operation data, and the acquired data is classified according to the transformer type. The prediction model of the same type of actual data is used for training. When predicting, the transformer type to be predicted is first distinguished, and the corresponding type of prediction model is used for oil temperature prediction to improve the prediction ability of the prediction model for actual operation data. An integrated learning model is used as the prediction model, and different models are used to capture different features in the data to improve the generalization ability and prediction accuracy of the model.

[0015] Furthermore, the oil temperature influencing parameters include ambient temperature, rated capacity, cooling method, load rate and current.

[0016] Furthermore, the actual operation data is preprocessed to fill missing values ​​and remove abnormal values, including:

[0017] Perform missing value detection on the actual operation data by using the notnull function, and fill the missing values ​​with the mean;

[0018] Perform duplicate value detection on the actual operation data by comparing adjacent data, and remove duplicate data;

[0019] The actual operating data is subjected to outlier detection through a box plot to eliminate abnormal data.

[0020] Furthermore, the performing outlier detection on the actual operation data by using a box plot includes:

[0021] Grouping the actual operation data according to data types, and drawing a box plot based on the grouped data;

[0022] The quartiles and interquartile ranges were calculated for each box, and the outlier limits were determined based on the quartiles and interquartile ranges;

[0023] Data outside the outlier limit are removed.

[0024] Furthermore, the preprocessing also includes: determining a normal data range in combination with actual operation and inspection experience, and eliminating data exceeding the normal data range.

[0025] Furthermore, the ensemble learning model is composed of a plurality of base learners and a meta learner;

[0026] The base learners include a multivariate linear regression model, a Ridge regression model, a support vector machine regression model, a decision tree regression model and a KNN regression model;

[0027] The meta-learner is a convolutional neural network based on the attention mechanism.

[0028] Furthermore, the convolutional neural network based on the attention mechanism includes a double-layer convolutional network, a gated recurrent network, an attention mechanism network and a fully connected layer connected in sequence;

[0029] The two-layer convolutional network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and a fully connected layer connected in sequence; the first convolutional layer and the second convolutional layer are one-dimensional convolutional layers, and the first pooling layer and the second pooling layer are maximum pooling layers.

[0030] The second aspect of the present invention provides a transformer oil temperature prediction device based on actual operation data, comprising:

[0031] A data acquisition module, used to acquire actual operation data of the transformer, wherein the actual operation data includes oil temperature and oil temperature influencing parameters;

[0032] A data processing module is used to classify the actual operation data into different transformer types according to the operating voltage level and cooling method of the transformer; and is also used to,

[0033] Preprocessing the actual operation data to fill missing values ​​and remove outliers;

[0034] A prediction module is used to construct an integrated learning model, and the integrated learning model is trained by the actual operation data of the same transformer type to obtain an oil temperature prediction model for each transformer type. The input of the integrated learning model is the oil temperature influencing parameter, and the output is the oil temperature; it is also used to,

[0035] The oil temperature of each type of transformer is predicted by the oil temperature prediction model of each type of transformer.

[0036] According to a third aspect of the present invention, an electronic device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for predicting transformer oil temperature based on actual operating data as described in any one of the first aspects of the present invention is implemented.

[0037] A third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the transformer oil temperature prediction method based on actual operating data as described in any one of the first aspect of the present invention.

[0038] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0039] The prediction model is trained based on the actual operating data of different types of transformers. When predicting the oil temperature, the type of transformer is first determined, and then a suitable oil temperature prediction model is selected. This improves the prediction ability of the prediction model in actual engineering applications and improves the prediction accuracy.

[0040] An integrated learning model is established, which utilizes the feature extraction capabilities of different models to improve the prediction capabilities of large-scale real data and the generalization performance of the model.

[0041] In data preprocessing, screening rules based on actual operation and inspection experience are established to improve the accuracy and standardization of data. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0043] Figure 1 is a flow chart of a transformer oil temperature prediction method based on actual operation data according to an embodiment of the present invention;

[0044] Figure 2 is a processing flow chart of a principal component analysis method according to an embodiment of the present invention;

[0045] Figure 3 is a schematic diagram of transformer type classification according to an embodiment of the present invention;

[0046] Figure 4 is a flow chart of a method for detecting outliers on actual operating data by using a box plot according to an embodiment of the present invention;

[0047] Figure 5 is a schematic diagram of a data preprocessing process according to an embodiment of the present invention;

[0048] Figure 6 is a flow chart of an embodiment of the present invention for predicting transformer oil temperature using a support vector machine regression prediction method;

[0049] Figure 7 This is a flowchart of using decision tree regression to predict transformer oil temperature in an embodiment of the present invention;

[0050] Figure 8 is a flow chart of predicting transformer oil temperature by KNN regression in an embodiment of the present invention;

[0051] Fig. 9 Schematic diagram of a convolutional neural network structure based on an attention mechanism according to an embodiment of the present invention;

[0052] Fig.10 is a schematic diagram of an oil temperature prediction process according to an embodiment of the present invention;

[0053] Fig.11 is a schematic diagram of an integrated learning model training process according to an embodiment of the present invention;

[0054] Fig.12 It is the loss function of the ensemble learning model trained with the 220kVONAN transformer dataset;

[0055] Fig.13 It is the loss function of the ensemble learning model trained with the 220kVONAN transformer dataset. DETAILED DESCRIPTION

[0056] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0057] It should be noted that the terms "including" and "having" and any variations thereof in the specification and claims of the present invention and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to or inherent to other steps or units of the device.

[0058] The terms used in the various embodiments of the application are only used to describe the purpose of specific embodiments and are not intended to limit the various embodiments of the application. As used herein, the singular form is intended to also include the plural form, unless the context clearly indicates otherwise. Unless otherwise limited, all terms used here (including technical terms and scientific terms) have the same meaning as the meanings commonly understood by ordinary technicians in the field of the various embodiments of the application. The terms (such as the terms defined in the dictionary generally used) will be interpreted as having the same meaning as the contextual meaning in the relevant technical field and will not be interpreted as having an idealized meaning or an overly formal meaning, unless clearly defined in the various embodiments of the application.

[0059] The embodiments of the present invention provide a transformer oil temperature prediction method and device based on actual operation data, which are suitable for predicting the oil temperature of the transformer during actual application, are conducive to obtaining accurate oil temperature prediction results, and provide accurate prediction data for power system maintenance and fault prevention.

[0060] See also Figure 1 The flowchart of the transformer oil temperature prediction method based on actual operation data is shown. The transformer oil temperature prediction method adopted by the present invention mainly includes the following steps.

[0061] S1, obtaining actual operation data of the transformer, the actual operation data including oil temperature and oil temperature influencing parameters.

[0062] S2, divides the actual operating data into different transformer types according to the operating voltage level and cooling method of the transformer.

[0063] S3, preprocess the actual operation data, fill in missing values ​​and remove outliers.

[0064] S4, constructing an integrated learning model, training the integrated learning model through actual operation data of the same transformer type, and obtaining an oil temperature prediction model for each transformer type. The input of the integrated learning model is the oil temperature influencing parameter, and the output is the oil temperature;

[0065] S5, predicting the oil temperature of each type of transformer by using the oil temperature prediction model of each transformer type.

[0066] The key oil temperature influencing parameters include the current ambient temperature, rated capacity, cooling method, load rate, current, etc. of the transformer to be predicted. There are many influencing parameters, and the key parameters can generally be screened out through principal component analysis or correlation analysis to extract features from the key parameters.

[0067] Principal component analysis is a statistical method used to discover patterns in data. It transforms a set of possibly correlated variables into linearly uncorrelated variables through orthogonal transformation. The transformed uncorrelated variables are called principal components. Figure 2 shown.

[0068] Principal component analysis (PCA) requires standardization of the original data to ensure that the mean of each eigenvalue is 0 and the standard deviation is 1. Considering that PCA is affected by the data scale, the data needs to be centered. The covariance matrix calculation is to calculate the covariance matrix or correlation coefficient matrix to determine the correlation between data features, where the covariance describes the correlation between two data. A value close to 1 indicates positive correlation, a value close to -1 indicates negative correlation, and a value close to 0 indicates no correlation. The formula for calculating the covariance of two data is as follows.

[0069]

[0070] Among them, X, represent input variables and output variables respectively. They represent the means of the input variables and output variables respectively, and n represents the number of variables.

[0071] After the covariance matrix is ​​calculated, the eigenvalues ​​need to be decomposed to obtain eigenvalues ​​and eigenvectors. The eigenvalues ​​represent the variance contribution of each principal component, while the eigenvectors represent the direction of the principal component. According to the size of the eigenvalues, the eigenvectors corresponding to the first few largest eigenvalues ​​are selected as the principal components. Finally, the original data set is projected onto these principal components to obtain a new data set, thus achieving data dimensionality reduction.

[0072] The present invention establishes data-driven transformer oil temperature prediction models for different category combinations, such as Figure 3 As shown, according to the operating voltage level of the transformer, it can be divided into 110kV, 220KV, 500KV, etc., and the cooling method can be divided into oil-immersed natural cooling (ONAN), oil-immersed air cooling (ONAF), and forced oil circulation air cooling (OFAF).

[0073] Before building the model, data preprocessing is performed first. The data preprocessing process includes data missing value detection and outlier inspection. The preprocessing steps are as follows:

[0074] The missing values ​​of the actual running data are detected by the notnull function, and the missing values ​​are filled with the mean value;

[0075] Perform duplicate value detection on actual operation data by comparing adjacent data and eliminate duplicate data;

[0076] The actual operation data is tested for outliers through box plots to eliminate abnormal data.

[0077] Considering that some data may be missing, for example, some substations cannot detect the ambient temperature, an intelligent supplement method for ambient temperature is set up. By obtaining the highest and lowest temperatures of the day in the area, assuming that the temperature is lowest at 2 a.m. and highest at 2 p.m., an arithmetic progression is designed based on the data sampling interval of the day to intelligently supplement the ambient temperature. The specific supplement process is as follows:

[0078]

[0079] Where, T Δ Represents the ambient temperature difference between two consecutive data points, T max Represents the highest temperature of the day, T min Represents the lowest temperature of the day, T i+1 is the ambient temperature value at the next moment, T i It is the current ambient temperature value.

[0080] Among them, the method of detecting outliers on actual operation data through box plots can be found in Figure 4 As shown, specifically including:

[0081] Group the actual operation data according to the data type, and draw a box plot based on the grouped data;

[0082] The quartiles and interquartile ranges were calculated for each box, and the outlier limits were determined based on the quartiles and interquartile ranges;

[0083] Eliminate data that exceed the outlier limit.

[0084] The box plot shows the distribution of the data, and then the quartiles and interquartile ranges are calculated to determine the outlier limits. The outliers that exceed the limits are marked on the box plot, and then data analysis and data verification are performed on the abnormal data to eliminate the abnormal data.

[0085] In terms of data preprocessing, most studies have performed missing value detection and some outlier detection based on error rules on transformer data. However, in the actual operation of the transformer, the anomaly of the data cannot be well detected by relying on the error rules, and the standardization and accuracy of the data set need to be verified. Based on this, the present invention proposes a method of proposing abnormal data in combination with actual operation and inspection experience, determines the normal data range based on the actual operation and inspection experience, and eliminates the data that exceeds the normal data range.

[0086] like Figure 5 The schematic diagram of the data preprocessing process shown in the figure performs missing value detection and duplicate value detection on the original data in turn, then performs outlier detection based on the box plot and operation and inspection experience, and finally selects key feature data through principal component analysis.

[0087] The outlier detection rules combined with actual operation and inspection experience are based on the criteria summarized from the actual operation and inspection projects of substations, such as the transformer oil temperature does not exceed 90 degrees, and the load rate mutation does not exceed 100% within a continuous time point. The box plot detection method combines actual operation and inspection experience to screen outliers to improve the integrity and accuracy of the data.

[0088] Ensemble learning methods improve the accuracy and robustness of predictions by combining multiple different models, which can reduce the deviation of single model predictions. Because ensemble learning methods can learn feature data from multiple models, they are more stable when dealing with noise and outliers. Ensemble learning methods flexibly select and adjust different base learners and integration strategies according to different application scenarios and requirements, and select effective feature and model combinations to improve prediction performance on large-scale data sets.

[0089] Furthermore, the integrated learning model consists of multiple base learners and a meta-learner. Taking the integrated learning structure of five base learners and one meta-learner as an example, five different prediction models are used to predict the oil temperature value of the transformer. The modeling process of each method is different, and the feature extraction method is also different.

[0090] An implementation of the present invention selects base learners as: multivariate linear regression model, Ridge regression model, support vector machine regression model, decision tree regression model and KNN regression model. The meta-learner adopts a convolutional neural network model based on the attention mechanism. The prediction methods of the five base models will be introduced respectively below.

[0091] 1. Multiple Linear Regression Model

[0092] The prediction method based on linear regression is a traditional technique used to predict the relationship between a continuous dependent variable and one or more predictor independent variables. It usually performs better when there is a good linear relationship between the input and output. Its formula is as follows:

[0093] f(x)=L1x1+L2x2+L3x3+L4x4++L5x5+b

[0094] Where f(x) represents the model output oil temperature, L represents the weight coefficient trained by the model, x1 is the model input, represents the transformer load rate, x2 represents the transformer cooling method, x3 represents the transformer rated capacity, x4 represents the transformer ambient temperature, x5 represents the current current of the transformer, and b is the deviation.

[0095] By training the linear regression model with large-scale data, the weights of different influencing factors can be obtained. The fitting process of multivariate linear regression usually uses the least squares method to determine the coefficients.

[0096] 2. Ridge regression model

[0097] Ridge regression, also known as ridge regression, is a linear regression method for processing collinear data. In multivariate linear regression, when there is multicollinearity in the predictor variables, ordinary least squares estimation will produce unstable results. Ridge regression introduces regularization terms to solve this problem. Rideg regression adds regularization terms to the loss function to shrink parameter estimates, thereby reducing model complexity. The weight of the regularization term is controlled by the ridge parameter λ, and the value of λ represents the degree of regularization. The following introduces the parameter derivation of Ridge regression.

[0098] The objective function of linear regression is:

[0099] J(β)=∑(y-Xβ) 2

[0100] Among them, β represents the linear regression coefficient, X represents the predicted value, and y represents the true value.

[0101] In order to ensure that the regression coefficient β can be calculated, the Ridge regression model adds an L2 norm penalty term to the objective function:

[0102] J(β)=∑(y-Xβ) 2 +∑λβ 2

[0103] Among them, λ is a non-negative number. The larger λ is, the smaller the regression coefficient β will be in order to minimize J(β).

[0104] The derivation process of the regression coefficient β is as follows:

[0105]

[0106] make

[0107]

[0108] The addition of the L2 norm penalty term makes the matrix full rank and ensures reversibility, but due to the addition of the penalty term, the estimation of the regression coefficient β is no longer an unbiased estimate. Regularization improves the stability of the model and reduces the risk of overfitting. There may be multicollinearity problems in the multi-source data of the substation. The introduction of regularization terms can reduce parameter fluctuations and estimate parameters more accurately. It has better prediction performance than multivariate linear regression.

[0109] 3. Support Vector Machine Regression Model

[0110] Support vector machine (SVM) is a supervised learning algorithm that can effectively process high-dimensional data and nonlinear relationships, and has a strong tolerance for noise and outliers. Therefore, it is widely used in regression prediction. The process of using support vector machine regression prediction method to predict transformer oil temperature is as follows: Figure 6 shown.

[0111] Data preprocessing is required before data standardization, because SVM is sensitive to data scale, and usually the data needs to be standardized or normalized. The core of the support vector machine method is to map low-dimensional space data to high-dimensional space through kernel functions, find a hyperplane in high-dimensional space, so that the error between the training sample and the hyperplane is minimized, and finally complete the regression prediction by weighted averaging the support vector through the kernel function. When the support vector machine performs nonlinear regression prediction, it is necessary to assign initial values ​​to the penalty factor C and the kernel parameter γ, but the determination of the initial value is related to the final prediction result of the model, so these parameters need to be optimized, and some bionic algorithms are usually used to optimize their hyperparameters. Finally, the training model is used to predict the oil temperature of the transformer.

[0112] 4. Decision Tree Regression Model

[0113] The decision tree is a basic classification and regression method, and the regression decision tree mainly refers to the CART algorithm. The value of its internal node features is yes or no, and it has a binary tree structure. The so-called regression is to determine the corresponding output value based on the feature vector. The regression tree divides the feature space into several units, and each division module has a specific output. The input of the algorithm is the training data set D, and the output is the regression tree f(x). The specific process of constructing the regression tree is: in the input space where the training data is located, recursively divide each area into two sub-areas and determine the output value of each sub-area to construct a binary decision tree. The process of using decision tree regression to predict transformer oil temperature is as follows: Figure 7 shown.

[0114] The specific process is to first input the feature data after data preprocessing, and then use the training set data to build a decision tree, usually starting from the root node, and recursively create child nodes based on the best segmentation point of the feature. When selecting the segmentation criterion, the decision tree algorithm needs to determine how to evaluate the feature segmentation effect. In the regression problem, the mean square error is the average value of the difference between the predicted value and the actual value. The segmentation criterion is to select the features and thresholds that can minimize the mean square error. The purpose is to find the optimal features for segmenting the data. Finally, model training and oil temperature prediction are performed. The details are as follows:

[0115] (1) Select the optimal segmentation variable j and segmentation point s and solve the formula:

[0116]

[0117] Traverse the variable j, scan the split point s for the fixed split variable j, and select (j, s) that minimizes the above equation.

[0118] (2) Use the selected (j, s) to divide the region and determine the corresponding output value:

[0119]

[0120] (3) Continue calling steps (1) and (2) for the two sub-regions until the stopping condition is met.

[0121] (4) Divide the input space into M regions R1, R2, ..., Rm and generate a decision tree:

[0122]

[0123] Where I is the indicator function:

[0124]

[0125] 5. KNN regression model

[0126] The KNN algorithm is a type of machine learning algorithm. Its core idea is: for any n-dimensional input vector, each corresponds to a point in the feature space, and the output is the category label or predicted value corresponding to the feature vector. The KNN algorithm does not have a learning process in the traditional sense. Its working principle is to use training data to divide the feature vector space and use the division result as the final algorithm model. The general steps for predicting transformer oil temperature through KNN regression are as follows: Figure 8 shown.

[0127] When using the KNN algorithm for regression prediction, we find the K nearest neighbors of the new prediction instance and then calculate the target value y of these k samples. i The mean can be used as the predicted value of the sample:

[0128]

[0129] When using KNN regression prediction, you need to choose a distance measurement method to calculate the similarity between data points. Commonly used data metrics include Euclidean distance and Manhattan distance. Here, we choose Euclidean distance to calculate data relevance. Euclidean distance is also called Euclidean distance or L2 norm distance, which is the straight-line distance between two points. On an n-dimensional plane, x 1k 、x 2k The formula for calculating the Euclidean distance between two points is:

[0130]

[0131] The K value is equivalent to the hyperparameter of the KNN algorithm. The number of neighbors K needs to be selected according to the data set and the problem. Then the distance metric is calculated, and the similarity of different data points is determined by calculating the Euclidean metric. The weight determination process is to assign weights to neighbors to influence the influence of different neighbors. Usually the weight is inversely proportional to the distance, that is, the closer the distance, the greater the influence on the prediction result. After the parameters of the KNN regression are determined, the average (or weighted average) of the target values ​​of the K neighbors is calculated as the prediction result.

[0132] The integrated learning model uses a convolutional neural network with an attention mechanism. The convolutional neural network is a deep learning algorithm inspired by the biological visual cortex mechanism and is good at processing grid-like topological structures.

[0133] In the present invention, the convolutional neural network based on the attention mechanism includes a double-layer convolutional network, a gated recurrent network, an attention mechanism network and a fully connected layer connected in sequence.

[0134] like Fig. 9 The diagram of the convolutional neural network structure based on the attention mechanism is shown in the figure. The CNN layer, as a convolutional layer, forms a hierarchical structure through the stacking of multiple convolution and pooling layers, so that it can extract data features from low to high levels. The network structure adopts the CNN-GRU structure, which extracts features through the CNN convolution layer, learns features through the GRU layer, and then inputs them into the attention mechanism layer for weight distribution calculation after learning, and finally outputs the oil temperature prediction result through the fully connected layer.

[0135] The following is a brief description of the process of completing the oil temperature prediction in the form of mathematical expression. The characteristic data is expressed in P i Indicates, the convolutional layer is represented by CNN, and the feature vector is represented by H c Indicates that the GRU layer is represented by GRU, and the GRU layer input is represented by h i The attention mechanism layer is represented by SA, and the fully connected layer is represented by FC.

[0136] The feature data is converted into a feature vector through the CNN layer:

[0137] H c =CNN(P i )

[0138] The feature vector is passed through the GRU layer to complete feature learning:

[0139] h i =GRU(H c,i ,H c,i+1 ),i∈[1,n]

[0140] The output of the GRU layer is weighted through the attention mechanism layer:

[0141] Si =SA(h i )

[0142] The input of the output layer is the output of the attention mechanism layer, and the output layer calculates the final prediction result through full connection:

[0143] y i =FC(S i )

[0144] The specific implementation process is as follows: first, the prediction results of the base model are preprocessed to filter out outlier data points, and the preprocessed data is used as the input of the fine prediction model. The processed data set is input to the CNN layer, and the CNN layer extracts features from the data. The feature extraction process is as follows:

[0145]

[0146] P1=max(C1)+b2

[0147]

[0148] P2=max(C2)+b4

[0149] H C =f(P2×W3+b5)=Sigmoid(P2×W3+b5)

[0150] In the above process, C1 and C2 are the outputs of convolutional layer 1 and convolutional layer 2 respectively, P1 and P2 are the outputs of pooling layer respectively, W1, W2, W3 are weight matrices, b is the bias, is the convolution operation, H C It is the output of the CNN layer. After the feature vector is output, it passes through the GRU layer to complete feature learning, and then passes through the attention mechanism layer to calculate the weight probability corresponding to different feature vectors according to the weight distribution principle. The attention mechanism selects the Soft Attention mechanism, and its weight calculation process is:

[0151] e t =utanh(wh i +b)

[0152]

[0153]

[0154] In the formula, e t represents the attention probability distribution value determined by the output vector of the GRU network layer at the tth moment, u and w are the attention weight coefficients, b is the bias coefficient, and s t is the output of the attention mechanism layer, and the activation function is the Sigmoid function. Finally, the fully connected layer is used to calculate the precise prediction result.

[0155] In the process of predicting oil temperature based on the above-built integrated learning model, a transformer category classifier is first established to classify transformers according to voltage level and cooling method; then oil temperature prediction is performed for each specific category. Fig.10 The schematic diagram of the oil temperature prediction process shown in FIG. 1 specifically includes:

[0156] Stpe1. Establish a transformer classifier, classify transformers according to voltage level and cooling method, and divide the collected transformer operation data into different training sets according to transformer type.

[0157] Step 2: Preprocess the original data. The preprocessing process mainly includes duplicate value detection and screening, missing value detection and completion, outlier detection and screening, and principal component analysis method to extract main features.

[0158] Step 3. Establish a base model prediction module, input the feature data obtained after data preprocessing into the base model for training. The base model methods select multiple linear regression, ridge regression, support vector regression, decision tree regression, and KNN regression respectively. Each base training model will obtain an oil temperature prediction result.

[0159] Step 4: Establish an integrated learning prediction module, use the oil temperature results predicted by the five base models as the input of the meta-learner, and output the prediction results.

[0160] The training process is as follows Fig.11 As shown in the figure, the oil temperature results predicted by the five models are first input into the convolutional neural network. The convolutional neural network structure adopts the CNN-GRU structure, in which the attention mechanism weights are assigned after the GRU feature learning. The input of the convolutional neural network model is the prediction results of the five base models, and the output of the model is the true value of the oil temperature. The CNN layer extracts features from the input sequence and constructs a CNN framework consisting of 2 convolutional layers, 2 pooling layers and a fully connected layer. According to the characteristics of the load data, convolutional layer 1 and convolutional layer 2 are designed as one-dimensional convolutions, and the ReLU activation function is selected for activation. In order to retain more data fluctuation information, the pooling method of pooling layer 1 and pooling layer 2 selects maximum pooling. The original data processed by the convolutional layer and the pooling layer is feature mapped, and then feature learning is performed by the GRU layer. The feature vector results after learning are input into the attention mechanism layer, and finally the prediction results are calculated by the fully connected layer.

[0161] In summary, a single model may perform well on a specific data set, but its generalization ability is limited under different environments or conditions. The present invention adopts an integrated learning model. Different machine learning models can capture different features in the data. Combined with the deep learning model, it can better handle complex nonlinear relationships and enhance the generalization ability of the model. In the data preprocessing process, the model input variable type is determined by principal component analysis, and then the data is tested for missing values ​​and outliers. The outlier detection is completed using the box plot method combined with actual engineering operation and inspection experience, which can improve the standardization and accuracy of the data set. Combined with the actual engineering situation, considering the missing ambient temperature of the substation, an intelligent supplement method for ambient temperature is proposed to make up for the missing data. The present invention predicts the oil temperature by means of integrated learning, and can establish a prediction model for a large-scale data set, effectively improve the prediction accuracy of the transformer oil temperature, and improve the generalization and robustness of the prediction model.

[0162] Furthermore, the present invention establishes an oil temperature prediction model to perform prediction on 220kvONAN transformers and 110kvONAF transformers. The 220kVONAN transformer data set has 5600 sets of data, and its training process loss function is as follows: Fig.12 As shown in Figure 2, the root mean square error (RMSE) of the prediction result is 2.6784563. The size of the 110kVONAF transformer dataset is 2400 groups, and the loss function of its training process is as follows: Fig.13 As shown, the root mean square error (RMSE) of the prediction results is 1.7634856.

[0163] The results obtained on these two sets of experimental data show that the model has good prediction effect and generalization ability on large-scale actual data sets. Table 1 shows the comparison table of experimental results, comparing the prediction results of different cooling methods and different data scales.

[0164] Table 1 Comparison of experimental results

[0165]

[0166] By comparing the prediction results of different data sets, it is shown that the integrated learning model can predict oil temperature for large-scale data sets, and has good prediction accuracy, which meets the prediction requirements of actual engineering data.

[0167] An embodiment of the present invention further provides a transformer oil temperature prediction device based on actual operation data, which is used to execute the transformer oil temperature prediction method based on actual operation data, comprising:

[0168] A data acquisition module, used to acquire actual operation data of the transformer, wherein the actual operation data includes oil temperature and oil temperature influencing parameters;

[0169] A data processing module is used to classify the actual operation data into different transformer types according to the operating voltage level and cooling method of the transformer; and is also used to,

[0170] Preprocessing the actual operation data to fill missing values ​​and remove outliers;

[0171] A prediction module is used to construct an integrated learning model, and the integrated learning model is trained by the actual operation data of the same transformer type to obtain an oil temperature prediction model for each transformer type. The input of the integrated learning model is the oil temperature influencing parameter, and the output is the oil temperature; it is also used to,

[0172] The oil temperature of each type of transformer is predicted by the oil temperature prediction model of each type of transformer.

[0173] An embodiment of the present invention further provides an electronic device, which includes a processor and a memory, and the number of processors may be one or more. The memory, as a computer-readable storage medium, can be used to store software programs, computer executable programs, and modules. The processor executes various functional applications and data processing of the electronic device by running the software programs, instructions, and modules stored in the memory, so as to implement the transformer oil temperature prediction method based on actual operation data of any of the above embodiments of the present invention.

[0174] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the electronic device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0175] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the transformer oil temperature prediction method based on actual operation data according to any embodiment of the present invention is implemented.

[0176] An embodiment of the present invention further provides a computer program product. When the computer program product is run on a computer, the computer is enabled to execute the transformer oil temperature prediction method based on actual operation data in any of the above embodiments of the present invention.

[0177] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A transformer oil temperature prediction method based on actual operation data, characterized in that: include: Acquire actual operation data of the transformer, wherein the actual operation data includes oil temperature and oil temperature influencing parameters; Classifying the actual operating data into different transformer types according to the operating voltage level and cooling method of the transformer; Preprocessing the actual operation data to fill missing values ​​and remove outliers; Constructing an integrated learning model, training the integrated learning model through actual operation data of the same transformer type, and obtaining an oil temperature prediction model for each transformer type, wherein the input of the integrated learning model is an oil temperature influencing parameter, and the output is the oil temperature; The oil temperature of each type of transformer is predicted by the oil temperature prediction model of each type of transformer.

2. The transformer oil temperature prediction method based on actual operation data according to claim 1 is characterized in that: The oil temperature influencing parameters include ambient temperature, rated capacity, cooling method, load rate and current.

3. The transformer oil temperature prediction method based on actual operation data according to claim 1 is characterized in that: Preprocessing the actual operation data to fill missing values ​​and remove outliers includes: Perform missing value detection on the actual operation data by using the notnull function, and fill the missing values ​​with the mean; Perform duplicate value detection on the actual operation data by comparing adjacent data, and remove duplicate data; The actual operating data is subjected to outlier detection through a box plot to eliminate abnormal data.

4. The transformer oil temperature prediction method based on actual operation data according to claim 3 is characterized in that: The performing outlier detection on the actual operation data by using a box plot includes: Grouping the actual operation data according to data types, and drawing a box plot based on the grouped data; The quartiles and interquartile ranges were calculated for each box, and the outlier limits were determined based on the quartiles and interquartile ranges; Data outside the outlier limit are removed.

5. The transformer oil temperature prediction method based on actual operation data according to claim 3 is characterized in that: The preprocessing further includes: determining a normal data range in combination with actual operation and inspection experience, and eliminating data that exceeds the normal data range.

6. The transformer oil temperature prediction method based on actual operation data according to claim 1 is characterized in that: The integrated learning model consists of multiple base learners and a meta learner; The base learners include a multivariate linear regression model, a Ridge regression model, a support vector machine regression model, a decision tree regression model and a KNN regression model; The meta-learner is a convolutional neural network based on the attention mechanism.

7. The transformer oil temperature prediction method based on actual operation data according to claim 6 is characterized in that: The convolutional neural network based on the attention mechanism includes a double-layer convolutional network, a gated recurrent network, an attention mechanism network and a fully connected layer connected in sequence; The two-layer convolutional network includes a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and a fully connected layer connected in sequence; the first convolutional layer and the second convolutional layer are one-dimensional convolutional layers, and the first pooling layer and the second pooling layer are maximum pooling layers.

8. A transformer oil temperature prediction device based on actual operation data, characterized in that: include: A data acquisition module, used to acquire actual operation data of the transformer, wherein the actual operation data includes oil temperature and oil temperature influencing parameters; A data processing module is used to classify the actual operation data into different transformer types according to the operating voltage level and cooling method of the transformer; and is also used to, Preprocessing the actual operation data to fill missing values ​​and remove outliers; A prediction module is used to construct an integrated learning model, and the integrated learning model is trained by the actual operation data of the same transformer type to obtain an oil temperature prediction model for each transformer type. The input of the integrated learning model is the oil temperature influencing parameter, and the output is the oil temperature; it is also used to, The oil temperature of each type of transformer is predicted by the oil temperature prediction model of each type of transformer.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the transformer oil temperature prediction method based on actual operation data described in any one of claims 1 to 8 is 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 transformer oil temperature prediction method based on actual operation data described in any one of claims 1 to 8 is implemented.

Images