A method for predicting dissolved gas in transformer oil based on single variable de-sequencing

Abstract

This invention discloses a method for predicting dissolved gases in transformer oil based on univariate time-series deserialization. First, time-series samples of dissolved gases from online monitoring of transformer oil are collected and preprocessed. Then, a sliding window technique is used to extract features from historical data of dissolved gases in the oil, transforming the gas time-series sample data into non-time-series feature matrix samples. Finally, a deep forest prediction model is constructed, using a multi-granularity scanning module to deeply mine information from the non-time-series feature matrix. Through continuous training with a cascaded forest module, accurate prediction of the dissolved gas content in the oil at future times is achieved. This invention can further reduce the dependence on specific time points or temporal relationships without disrupting the "past predicts future" rule of time series, and improves the applicability of traditional non-time-series prediction models, thus contributing to improved prediction accuracy of online monitoring data of dissolved gases in transformer oil.

Description

Technical Field

[0001] This invention relates to the field of power equipment monitoring technology, specifically to a method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis. Background Technology

[0002] Online monitoring of dissolved gases in transformer oil is a crucial aspect of transformer equipment condition monitoring and maintenance. Accurate prediction of dissolved gases in transformer oil allows for early assessment of the transformer's operating status and timely detection of internal defects, significantly improving equipment reliability.

[0003] The current online monitoring data prediction methods for dissolved gases in transformer oil are mainly based on time series analysis, such as long short-term memory networks, moving average models, and convolutional neural networks. The principle is to predict the numerical trend of future time points by analyzing and modeling historical time series data. The above methods have achieved certain results in improving the prediction accuracy of dissolved gases in transformer oil, but there are still many problems to be solved in the field application of time series prediction methods, mainly including: (1) The generation and release of dissolved gases in transformer oil are affected by multiple factors, including operating load, external environmental interference, etc., which makes the construction of time series prediction models too complex and uncertain. (2) The changes in dissolved gases in transformer oil may have a time lag before the occurrence of internal defects in the equipment, and the prediction model is prone to ignoring this time lag effect, resulting in the inability to accurately capture the correlation between dissolved gases in oil and equipment status. (3) Compared with traditional prediction models, deploying and maintaining highly complex time series prediction models will increase the cost of system maintenance, and there are also problems such as over-reliance on specific time points or time series relationships. Summary of the Invention

[0004] To address the problems existing in the prior art, the purpose of this invention is to provide a method for predicting dissolved gases in transformer oil based on univariate de-time-series processing. This method performs de-time-series processing on online monitoring data of dissolved gases in transformer oil and achieves accurate prediction of dissolved gases in oil through a deep forest prediction model.

[0005] To achieve this objective, the present invention employs the following technical solution: a method for predicting dissolved gases in transformer oil based on univariate de-time-sequencing, comprising the following steps:

[0006] S1: Collect online monitoring data of dissolved gases in transformer oil, including 7 characteristic gases such as H2, CH4, C2H6, C2H4, C2H2, CO and CO2, and perform outlier processing and missing data filling on the time series sample data of the above characteristic gases;

[0007] S2: Use sliding window technology to extract features from historical data of dissolved gases in oil, and transform time series sample data into non-time series feature matrix samples;

[0008] S3: Divide the non-temporal feature matrix samples into training and test sets, construct a deep forest prediction model, and use the tree structure probability density estimation algorithm to optimize the parameters of the deep forest prediction model;

[0009] S4: Evaluate the performance of the deep forest prediction model, predict the dissolved gas content in oil at future moments, and optimize the operation and maintenance strategy of transformers in advance.

[0010] Furthermore, the outlier handling described in S1 uses box plots to identify and process outliers in the time-series sample data of dissolved gases in transformer oil. The processing rules are as follows:

[0011] ,

[0012] In the formula, T represents outliers in the time series sample data, and Q... u and Q L These are the upper quartile and lower quartile, Q. c For Q u and Q L The numerical difference.

[0013] Furthermore, the missing data imputation described in S1 involves using the KNN interpolation method based on nearest neighbor points to impute missing data in the time series sample data of dissolved gases in transformer oil. The specific steps are as follows:

[0014] S1.1: Use cross-validation to select the number of neighboring points, K.

[0015] S1.2: Calculate the distance between the target point and all other points using Euclidean distance. The calculation formula is as follows:

[0016] ,

[0017] In the formula, (x i , y i Let be the coordinates of the neighboring points and the target point in the i-th dimension, and m be the data dimension;

[0018] S1.3: Select the K data points closest to the target point, and assign weights to each neighboring point inversely proportional to the Euclidean distance. Calculate the weighted average of all neighboring points as the imputation value for missing data.

[0019] Furthermore, in S2, a sliding window technique is used to extract features from historical data of dissolved gases in oil, transforming the time-series sample data into a non-time-series feature matrix. The specific steps are as follows:

[0020] S2.1: Determine the sliding window size and step size based on the time series sample data of dissolved gases in transformer oil;

[0021] S2.2: Perform "sliding window" processing on the time series samples of each characteristic gas and calculate the statistical characteristics of the gas in each window, including the mean, maximum, minimum and standard deviation;

[0022] S2.3: Integrate the features extracted from each window to generate a feature matrix, where each row represents the features extracted from a time window and each column represents a feature.

[0023] Furthermore, the deep forest prediction model is constructed in S3, and the specific steps are as follows:

[0024] S3.1: Based on the multi-granularity scanning module, feature extraction is performed on the input non-temporal feature matrix samples. Assuming the feature dimension is l, the scanning size is q, and the step size is k, the multi-granularity scanning module extracts a (lq) / k+1 dimensional sub-feature vector, which is used as input to the two evaluators, Random Forest and Fully Random Forest, for training. Each evaluator outputs an α-dimensional probability vector. All output vectors are concatenated to obtain the enhanced feature vector with the following dimension:

[0025] ,

[0026] S3.2: The feature vectors obtained from the multi-granularity scanning module are used as input to the cascaded forest, passed and trained step by step in each cascaded layer, and the predicted probability vectors and input feature vectors are concatenated and combined; the coefficient of determination R is used to... 2 The evaluation index for judging the expansion of cascading forests is calculated using the following formula:

[0027] ,

[0028] In the formula, n is the number of prediction points; For gas prediction values; y i This represents the actual gas value; The mean is the average value; the coefficient of determination is R. 2 The closer the value is to 1, the better the deep forest prediction model fits the input feature data;

[0029] S3.3: If If the cascaded forest expansion stops, the deep forest prediction model will output the final predicted value of dissolved gases in transformer oil.

[0030] Furthermore, in S3, a tree-structured probability density estimation algorithm is used to optimize the parameters of the deep forest prediction model. The specific steps are as follows:

[0031] S3-1: Determine the parameter space of the deep forest prediction model, including the number of trees, the depth of the trees, and the minimum number of samples per leaf node;

[0032] S3-2: The expected value enhancement method is used to measure the extent to which the loss value corresponding to each selected parameter can be reduced. The formula for calculating the expected value enhancement is as follows:

[0033] ,

[0034] In the formula, V represents the conditional probability distribution; v is the sampled value; v′ is the optimal sampling threshold; u is any value in the configuration space parameters;

[0035] S3-3: Select the parameter combination with the highest probability density and calculate the maximum expected value of the improvement as a candidate for the next iteration;

[0036] S3-4: Repeat steps S3-1 to S3-3 until the set number of iterations is reached or the stopping condition is met.

[0037] Furthermore, in S4, the performance of the deep forest prediction model is evaluated using evaluation metrics including the coefficient of determination R. 2 and root mean square error R MSE , where R MSE The calculation formula is as follows:

[0038] ,

[0039] By using a deep forest prediction model to predict the dissolved gas content in oil at future moments, the goal of optimizing transformer operation and maintenance strategies in advance can be achieved.

[0040] The beneficial effects of this invention are:

[0041] This invention provides a method for predicting dissolved gases in transformer oil based on univariate de-time-series processing. By de-time-series processing the online monitoring data of dissolved gases in transformer oil, the complexity of the prediction model can be reduced and the applicability of traditional non-time-series prediction models can be improved. This helps to improve the prediction accuracy and effect of online monitoring data of dissolved gases in transformer oil, and also provides a reliable guarantee for early fault warning of transformers. Attached Figure Description

[0042] Figure 1 This is a flowchart of the method for predicting dissolved gases in transformer oil based on univariate de-temporalization in this invention;

[0043] Figure 2 This is a schematic diagram of the de-temporalization of dissolved gases in oil in this invention;

[0044] Figure 3 This is the cascaded forest structure diagram in this invention;

[0045] Figure 4 This is the predicted result of methane gas in the embodiments of the present invention. Detailed Implementation

[0046] The present invention will be further described below with reference to the embodiments. It should be noted that the following embodiments are only used to further illustrate the present invention and should not be construed as limiting the scope of protection of the present invention. Some non-essential improvements and adjustments made by those skilled in the art based on the above-described invention are still within the scope of protection of the present invention.

[0047] like Figure 1 As shown, a specific embodiment of the present invention provides a method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis, comprising the following steps:

[0048] S1: Collect online monitoring data of dissolved gases in transformer oil, mainly including 7 characteristic gases such as H2, CH4, C2H6, C2H4, C2H2, CO and CO2. Perform outlier processing and missing data filling on the time series sample data of the above characteristic gases.

[0049] In this invention, outlier processing is performed on the time-series sample data of dissolved gases in transformer oil. Box plots are primarily used to identify and process outliers in the time-series sample data of dissolved gases in transformer oil. The processing rules are as follows:

[0050] ,

[0051] In the formula, T represents outliers in the time series sample data, and Q... u and Q L These are the upper quartile and lower quartile, Q. c For Q u and Q L The numerical difference. Data imputation for missing dissolved gases in transformer oil is primarily based on the KNN interpolation method for nearest neighbor points. The specific steps are as follows:

[0052] S1.1: Use cross-validation to select an appropriate number of neighboring points, K.

[0053] S1.2: Calculate the distance between the target point and all other points using Euclidean distance. The calculation formula is as follows:

[0054] ,

[0055] In the formula, (x i , y i) represents the coordinates of the neighboring point and the target point in the i-th dimension, and m is the data dimension.

[0056] S1.3: Select the K data points closest to the target point, and assign weights to each neighboring point inversely proportional to the Euclidean distance. Calculate the weighted average of all neighboring points as the imputation value for missing data.

[0057] S2: A sliding window technique is used to extract features from historical data of dissolved gases in transformer oil, transforming the time-series sample data into a non-time-series feature matrix sample. Univariate de-time-series transformation of dissolved gases in transformer oil is as follows: Figure 2 As shown, the specific steps are as follows:

[0058] S2.1: Determine the sliding window size and step size based on the time series sample data of dissolved gases in transformer oil.

[0059] S2.2: Perform sliding window processing on the time series samples of each characteristic gas and calculate the statistical characteristics of the gas in each window, mainly including the mean, maximum, minimum and standard deviation.

[0060] S2.3: Integrate the features extracted from each window to generate a feature matrix, where each row represents the features extracted from a time window and each column represents a feature.

[0061] S3: Divide the non-temporal feature matrix samples into training and test sets, construct a deep forest prediction model, and use a tree structure probability density estimation algorithm to optimize the parameters of the deep forest prediction model.

[0062] In this invention, a multi-granularity scanning module is used to extract features from the input non-temporal feature matrix samples. Assuming the feature dimension is l, the scanning size is q, and the step size is k, the multi-granularity scanning module can extract (lq) / k+1 dimensional sub-feature vectors, which are then used as inputs to two evaluators, Random Forest and Fully Random Forest, for training. Each evaluator outputs an α-dimensional probability vector. All output vectors are concatenated to obtain an enhanced feature vector with the following dimension:

[0063] ,

[0064] The feature vectors obtained from the multi-granularity scanning module are used as input to the cascaded forest, and the cascaded forest structure is as follows: Figure 3 As shown. Training is performed sequentially within each cascaded layer, and the predicted probability vector and the input feature vector are concatenated and combined. The coefficient of determination R0 is then used to determine the probability vector. 2 The evaluation index for judging the expansion of cascading forests is calculated using the following formula:

[0065] ,

[0066] In the formula, n is the number of prediction points; For gas prediction values; y i This represents the actual gas value; The average value is given. The coefficient of determination R0 2 The closer the value is to 1, the better the deep forest prediction model fits the input feature data. If the cascaded forest expansion stops, the deep forest prediction model will output the final predicted value of dissolved gases in transformer oil.

[0067] A tree-structure probability density estimation algorithm is used to optimize the parameters of the deep forest prediction model. First, the parameter space of the deep forest prediction model is determined, including hyperparameters such as the number of trees, tree depth, and minimum number of samples per leaf node. Then, expectation boosting is used to measure the reduction in loss value corresponding to each selected parameter. The expectation boosting formula is as follows:

[0068] ,

[0069] In the formula, Let v be a conditional probability distribution; v be a sampled value; v′ be the optimal sampling threshold; and u be any value in the configuration space parameters. Select the parameter combination with the highest probability density and calculate the maximum expected value as a candidate for the next iteration. Repeat the above steps until the set number of iterations is reached or the stopping condition is met.

[0070] S4: Evaluate the performance of the deep forest prediction model using evaluation metrics including the coefficient of determination R. 2 and root mean square error R MSE , where R MSE The calculation formula is as follows:

[0071] ,

[0072] By using a deep forest prediction model to predict the dissolved gas content in oil at future moments, the goal of optimizing transformer operation and maintenance strategies in advance can be achieved.

[0073] Example

[0074] The online monitoring interval for dissolved gases in transformer oil is 2 hours. Historical methane gas data from the past 30 days was extracted to predict the gas change trend for the next three days. A sliding window size of 12 and a sliding step of 1 were set, and the online methane monitoring data was de-time-series processed. The mean, maximum, minimum, and standard deviation of each sliding window were calculated as statistical features of methane gas content changes within a day. The data features and statistical features obtained from the sliding window were concatenated and concatenated, and samples with missing data were deleted, resulting in a 348×15 non-time-series feature matrix sample. The non-time-series feature matrix sample was divided into training and testing sets in a 7:3 ratio, and a deep forest prediction model was trained. A tree structure probability density estimation algorithm was used to simultaneously optimize the main parameters of the deep forest prediction model, including the number of trees, the maximum depth of the trees, and the minimum number of samples per leaf node, finally obtaining the optimal deep forest prediction model. The prediction results of the optimized deep forest prediction model for methane gas content in the next three days are as follows. Figure 4 As shown, its coefficient of determination is 0.87, the root mean square error is 48.2, and the predicted value and the true value are basically consistent, indicating that the present invention can achieve the prediction accuracy of dissolved gas in transformer oil and effectively improve the prediction effect of traditional machine learning regression model.

[0075] The above description merely illustrates preferred embodiments of the present invention, and while the description is relatively specific and detailed, it should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications, improvements, and substitutions without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this patent should be determined by the appended claims.

Claims

1. A method for predicting dissolved gases in transformer oil based on univariate de-temporalization, characterized in that: Includes the following steps: S1: Collect online monitoring data of dissolved gases in transformer oil, including 7 characteristic gases such as H2, CH4, C2H6, C2H4, C2H2, CO and CO2, and perform outlier processing and missing data filling on the time series sample data of the above characteristic gases; S2: Use sliding window technology to extract features from historical data of dissolved gases in oil, and transform time series sample data into non-time series feature matrix samples; S3: Divide the non-temporal feature matrix samples into training and test sets, construct a deep forest prediction model, and use the tree structure probability density estimation algorithm to optimize the parameters of the deep forest prediction model; S4: Evaluate the performance of the deep forest prediction model, predict the dissolved gas content in oil at future moments, and optimize the operation and maintenance strategy of transformers in advance. S2 uses a sliding window technique to extract features from historical data of dissolved gases in oil, transforming time-series sample data into a non-time-series feature matrix. The specific steps are as follows: S2.1: Determine the sliding window size and step size based on the time series sample data of dissolved gases in transformer oil; S2.2: Perform sliding window processing on the time series samples of each characteristic gas and calculate the statistical characteristics of the gas in each window, including the mean, maximum, minimum and standard deviation; S2.3: Integrate the features extracted from each window to generate a feature matrix, where each row represents the features extracted from a time window and each column represents a feature.

2. The method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis as described in claim 1, characterized in that: The outlier handling described in S1 uses box plots to identify and process outliers in the time-series sample data of dissolved gases in transformer oil. The processing rules are as follows: ， In the formula, T These are outliers in the time series sample data. Q u and Q L These are the upper quartile and the lower quartile, respectively. Q c for Q u and Q L The numerical difference.

3. The method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis as described in claim 1, characterized in that: The missing data imputation described in S1 involves imputing missing data in the time series sample data of dissolved gases in transformer oil based on the KNN interpolation method of nearest neighbors. The specific steps are as follows: S1.1: Use cross-validation to select the number of neighboring points. K value; S1.2: Calculate the distance between the target point and all other points using Euclidean distance. The calculation formula is as follows: ， In the formula, ( x i , y i ) for the neighboring points and the target point in the th i Coordinates in each dimension m For data dimensions; S1.3: Select the point closest to the target point K We have 100 data points, and we assign weights to each neighboring point inversely proportional to the Euclidean distance. We then calculate the weighted average of all neighboring points as the imputation value for missing data.

4. The method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis as described in claim 1, characterized in that: The specific steps for building a deep forest prediction model in S3 are as follows: S3.1: Feature extraction is performed on the input non-temporal feature matrix samples based on the multi-granularity scanning module, assuming the feature dimension is... l The scanning size is q Step size is k The multi-granularity scanning module extracts ( l - q ) / k +1 dimensional sub-feature vectors are used as inputs to train two evaluators: Random Forest and Fully Random Forest. Each evaluator outputs... α A probability vector of dimension ; concatenating all output vectors yields an enhanced feature vector of dimension . ， S3.2: The feature vectors obtained from the multi-granularity scanning module are used as input to the cascaded forest, passed and trained step by step in each cascaded layer, and the predicted probability vectors and input feature vectors are concatenated and combined; the coefficient of determination is used to... R 2 The evaluation index for judging the expansion of cascading forests is calculated using the following formula: ， In the formula, n The number of prediction points; This is a gas prediction value; y i This represents the actual gas value; The average value; coefficient of determination R 2 The value is less than or equal to 1. R 2 The larger the value, the better the deep forest prediction model fits the input feature data; S3.3: If If the cascaded forest expansion stops, the deep forest prediction model will output the final predicted value of dissolved gases in transformer oil.

5. The method for predicting dissolved gases in transformer oil based on univariate de-time-series analysis as described in claim 1, characterized in that: In S3, a tree-structured probability density estimation algorithm is used to optimize the parameters of the deep forest prediction model. The specific steps are as follows: S3-1: Determine the parameter space of the deep forest prediction model, including the number of trees, the depth of the trees, and the minimum number of samples per leaf node; S3-2: The expected value enhancement method is used to measure the extent to which the loss value corresponding to each selected parameter can be reduced. The formula for calculating the expected value enhancement is as follows: ， In the formula, It is a conditional probability distribution; v These are sampled values; v ′ is the optimal sampling threshold; u Allow any value in the configuration space parameters; S3-3: Select the parameter combination with the highest probability density and calculate the maximum expected value of the improvement as a candidate for the next iteration; S3-4: Repeat steps S3-1 to S3-3 until the set number of iterations is reached or the stopping condition is met.

6. The method for predicting dissolved gases in transformer oil based on univariate de-temporalization as described in claim 4, characterized in that: In S4, the performance of deep forest prediction models is evaluated using metrics including the coefficient of determination. R 2 and root mean square error R MSE ,in R MSE The calculation formula is as follows: ， By using a deep forest prediction model to predict the dissolved gas content in oil at future moments, the goal of optimizing transformer operation and maintenance strategies in advance can be achieved.

Images