Current transformer error state prediction method and system based on feature decoupling, storage medium and computer device

Abstract

The application discloses a current transformer error state prediction method and system based on feature decoupling, a storage medium and computer equipment. The method first collects historical secondary output current data of a three-phase current transformer and performs stationary treatment; features of adjacent data after stationary treatment are decoupled, static features and dynamic features are extracted, and the dynamic features are decomposed into dynamic trend features, dynamic period features and dynamic residual features; the feature contribution degree method is used to select optimal static features, optimal dynamic trend features, optimal dynamic period features and optimal dynamic residual features; and the optimal features are weighted and fused and then input into an LSTM model to obtain the error state of the current transformer. The application effectively solves the problems of large fluctuation of current time series data and insufficient feature extraction capability, improves the prediction accuracy of the error state of the current transformer, and is suitable for intelligent substation power metering online monitoring and equipment predictive maintenance scenarios.

Description

Technical Field

[0001] This invention relates to smart grid technology, specifically to a method, system, storage medium, and computer equipment for predicting the error state of a current transformer based on feature decoupling. Background Technology

[0002] A current transformer (CT) is a general-purpose device that converts a large current into a smaller current for measurement, and it is widely used in power systems. Metering performance is the most critical attribute of a current transformer. With long-term operation, current transformers are affected by various physical fields such as electricity, magnetism, heat, and force, which can cause error deviations or even abnormalities, leading to current measurement failures and seriously affecting the stable operation of the power system.

[0003] Currently, current transformer error assessment typically employs offline or online verification methods, obtaining the ratio error and phase angle error through direct comparison. However, these methods suffer from long verification cycles, complex field wiring, and low efficiency. To improve the current transformer error status assessment system, it is urgent to establish a current transformer error status prediction method to promptly detect error exceeding tolerances, reduce the operating time of current transformers exceeding error limits, guide transformer testing work, and thus ensure the fairness of electricity metering.

[0004] Currently, there is a certain research foundation for current transformer error state prediction methods based on artificial intelligence algorithms. However, prediction based on current time series data has many problems: on the one hand, the current time series data varies from 0% to 120% of the rated current, with huge fluctuations; on the other hand, the model cannot efficiently mine the deep relationship between the features of the current time series data, making it difficult to make accurate predictions. Summary of the Invention

[0005] To overcome the aforementioned shortcomings and deficiencies of existing technologies, the primary objective of this invention is to provide a current transformer error state prediction method based on feature decoupling. This method solves the problems of large fluctuations in current transformer time-series data and low model prediction accuracy, thereby achieving accurate prediction of current transformer errors.

[0006] The second objective of this invention is to provide a current transformer error state prediction system based on feature decoupling.

[0007] A third objective of this invention is to provide a storage medium.

[0008] A fourth objective of this invention is to provide a computer device.

[0009] The objective of this invention is achieved through the following technical solution: This invention provides a method for predicting the error state of a current transformer based on feature decoupling, comprising the following steps:

[0010] S1. Collect historical secondary output current data of a three-phase current transformer operating on a specific line in a substation. And this historical secondary output current data Stabilization process;

[0011] S2. The adjacent data obtained after stabilization processing , and Feature decoupling is performed to obtain static and dynamic features; then the dynamic features are decomposed to obtain dynamic trend features, dynamic periodic features, and dynamic residual features.

[0012] S3. Use the feature contribution method to determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature;

[0013] S4. The optimal static feature, optimal dynamic trend feature, optimal dynamic periodic feature and optimal dynamic residual feature are fused to obtain the fused feature;

[0014] S5. Input the fused features into the LSTM to obtain the error of the current transformer.

[0015] Preferably, the stabilization process in step S1 is as follows:

[0016]

[0017]

[0018]

[0019] Where i represents the number of current transformers, and ω represents the number of sampling points. This represents the average value of historical secondary output current data. This represents the variance of the historical secondary output current data. This represents the historical secondary output current data after stabilization. This represents the cross product.

[0020] Preferably, step S2 includes the following specific steps:

[0021] S21. Extract the first... using a dynamic encoder and a static encoder respectively. , , Historical secondary output current data after time stabilization , , Dynamic and static characteristics;

[0022] S22. The STL algorithm is used to decompose the dynamic features to obtain dynamic trend features, dynamic periodic features, and dynamic residual features:

[0023]

[0024]

[0025]

[0026] in, , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics.

[0027] Preferably, the extraction process of dynamic and static features in step S21 is as follows:

[0028] Both the dynamic and static encoders employ five layers of 4×4 convolutional operations: batch normalization and Leaky ReLU activation are applied after the first four convolutional layers, while the output layer uses the Tanh activation function to normalize the output motion feature vector to the range of -1 to 1.

[0029]

[0030]

[0031]

[0032]

[0033]

[0034]

[0035] in, , , They represent , , Dynamic characteristics; , , They represent , , Static characteristics.

[0036] Preferably, the specific process of step S3 is as follows:

[0037]

[0038] in, Let j be the contribution level, and j be the j-th feature. For the first There are features, where P is the total number of features. For the generalized function of the LSTM algorithm, The average of all predicted feature values. It is a set that includes all dynamic and static features.

[0039] Based on the calculation results, the feature with the largest contribution is selected as the optimal feature.

[0040] Preferably, the specific process of step S4 is as follows:

[0041]

[0042]

[0043]

[0044] in, The symbol represents the activation function, Avg represents the average pooling function, and Cov represents the convolution function. =1, 2, 3, 4, Indicates the first The weighted weights of each feature, This represents the weights after fusion.

[0045] Preferably, the specific process of step S5 is as follows:

[0046]

[0047] Where Y represents the error state of the current transformer, and lstm is the generalized function of the LSTM algorithm.

[0048] A current transformer error state prediction system based on feature decoupling includes:

[0049] The data acquisition and preprocessing module is used to acquire historical secondary output current data of a three-phase current transformer in a substation during operation, and to perform stabilization processing on this historical secondary output current data.

[0050] The feature extraction module is used to decouple the adjacent data obtained after stabilization to extract static and dynamic features, and to decompose the dynamic features to extract dynamic trend features, dynamic periodic features and dynamic residual features.

[0051] The fusion module is used to determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature, and to fuse these features to obtain the fused feature.

[0052] The output module uses LSTM to process the fused features to obtain the error of the current transformer.

[0053] A storage medium stores a program, which, when executed by a processor, achieves the first objective of the current transformer error state prediction method based on feature decoupling.

[0054] A computer device, including a processor and a memory for storing processor-executable programs, characterized in that, when the processor executes the program stored in the memory, it implements the current transformer error state prediction method based on feature decoupling as described in the first objective.

[0055] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0056] 1. This invention performs stabilization processing on historical secondary output current data to eliminate mean bias and scale differences in the original data, providing stable and standardized input for subsequent feature extraction. This effectively solves the problem of large fluctuations in current time series data, thereby improving the model convergence speed and training stability.

[0057] 2. This invention uses a dynamic encoder and a static encoder to extract the dynamic and static features of current data respectively, and combines the STL algorithm to further decompose the dynamic features into three components: trend, period, and residual. This can fully explore the feature information at different levels in time series data, improve the model's ability to understand complex change patterns, and solve the problem of insufficient feature extraction capability of existing methods.

[0058] 3. This invention uses the feature contribution method to quantitatively evaluate each feature, automatically selects the optimal feature that contributes the most to the error state, avoids interference from redundant features, improves the generalization ability and prediction efficiency of the model, and achieves dimensionality reduction and enhancement at the feature level.

[0059] 4. This invention, by weightedly fusing the optimal static features, dynamic trend features, dynamic periodic features, and dynamic residual features, and combining the fused features with an LSTM model, can more accurately capture the changing patterns of the error state of the current transformer, achieve precise prediction of the three states of normal, abnormal, and alarm, and significantly improve the accuracy of error state classification.

[0060] 5. Based on this method, the prediction system, storage medium and computer equipment are deployed. This invention has high engineering practical value and can be widely applied to scenarios such as smart substations and online monitoring of power metering, providing technical support for the condition assessment and health management of current transformers. Attached Figure Description

[0061] Figure 1 This is a flowchart of a current transformer error state prediction method based on feature decoupling, according to an embodiment of the present invention.

[0062] Figure 2 This is a flowchart illustrating the feature extraction process according to an embodiment of the present invention. Detailed Implementation

[0063] The present invention is further described below through specific embodiments, but the scope of protection of the present invention is not limited thereto.

[0064] like Figure 1 As shown, the current transformer error state prediction method based on feature decoupling includes the following steps:

[0065] S1. Collect historical secondary output current data of a three-phase current transformer operating on a specific line in a substation. And this historical secondary output current data A stabilization process is performed; the stabilization process in step S1 is as follows:

[0066]

[0067]

[0068]

[0069] Where i represents the number of current transformers, and in this embodiment, data is collected from three current transformers, i.e., i = 1, 2, 3, and ω represents the number of sampling points. This represents the average value of historical secondary output current data. This represents the variance of the historical secondary output current data. This represents the historical secondary output current data after stabilization, and the cross product is also represented.

[0070] like Figure 2 As shown, S2 represents the adjacent data obtained after stabilization processing. , and Feature decoupling is performed to obtain static and dynamic features; then, the STL (Seasonal-Trend decomposition of Time Series) algorithm is used to decompose the dynamic features to obtain dynamic trend features, dynamic periodic features, and dynamic residual features; step S2 includes the following specific steps:

[0071] S21. Employs a dynamic encoder. and static encoder Extract the first one respectively , , Historical secondary output current data after time stabilization , , Dynamic and static characteristics;

[0072] S22. The STL algorithm is used to decompose the dynamic features to obtain dynamic trend features, dynamic periodic features, and dynamic residual features:

[0073]

[0074]

[0075]

[0076] in, , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics.

[0077] The extraction process of dynamic and static features in step S21 is as follows:

[0078] Both the dynamic and static encoders employ five layers of 4×4 convolutional operations: batch normalization and Leaky ReLU activation are applied after the first four convolutional layers, while the output layer uses the Tanh activation function to normalize the output motion feature vector to the range of -1 to 1.

[0079]

[0080]

[0081]

[0082]

[0083]

[0084]

[0085] in, , , They represent , , Dynamic characteristics; , , They represent , , Static characteristics.

[0086] S3. Determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature using the feature contribution method; the specific process of step S3 is as follows:

[0087] The specific process of step S3 is as follows:

[0088]

[0089] in, Let j be the contribution level, and j be the j-th feature. For the first There are features, where P is the total number of features. For the generalized function of the LSTM algorithm, The average of all predicted feature values. It is a set that includes all dynamic and static features.

[0090] Based on the calculation results, the feature with the largest contribution is selected as the optimal feature.

[0091] S4. The optimal static feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature are fused to obtain the fused feature; the specific process of step S4 is as follows:

[0092]

[0093]

[0094]

[0095] in, The symbol represents the activation function, Avg represents the average pooling function, and Cov represents the convolution function. =1, 2, 3, 4, Indicates the first The weighted weights of each feature, This represents the weights after fusion.

[0096] S5. Input the fused features into an LSTM (Long Short-Term Memory Neural Network) to obtain the error of the current transformer.

[0097] The specific process of step S5 is as follows:

[0098]

[0099] Where Y represents the error state of the current transformer, and lstm is the generalized function of the LSTM algorithm. Specifically, the error state is divided into three types: normal, abnormal, and alarm.

[0100] A current transformer error state prediction system based on feature decoupling includes:

[0101] The data acquisition and preprocessing module is used to acquire historical secondary output current data of a three-phase current transformer in a substation during operation, and to perform stabilization processing on this historical secondary output current data.

[0102] The feature extraction module is used to decouple the adjacent data obtained after stabilization to extract static and dynamic features, and to decompose the dynamic features to extract dynamic trend features, dynamic periodic features and dynamic residual features.

[0103] The fusion module is used to determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature, and to fuse these features to obtain the fused feature.

[0104] The output module uses LSTM to process the fused features to obtain the error of the current transformer.

[0105] The storage medium stores a program, which, when executed by a processor, implements the current transformer error state prediction method based on feature decoupling described in this invention.

[0106] A computer device, including a processor and a memory for storing processor-executable programs, characterized in that, when the processor executes the program stored in the memory, it implements the current transformer error state prediction method based on feature decoupling described in this invention.

[0107] Those skilled in the art will readily understand that the above description is merely an embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A current transformer error state prediction method based on feature decoupling, characterized in that, Includes the following steps: S1. Collect historical secondary output current data of a three-phase current transformer operating on a specific line in a substation. And this historical secondary output current data Stabilization process; S2. The adjacent data obtained after stabilization processing , and Feature decoupling is performed to obtain static and dynamic features; The dynamic features are further decomposed to obtain dynamic trend features, dynamic periodic features, and dynamic residual features; Step S2 includes the following specific steps: S21. Extract the first... using a dynamic encoder and a static encoder respectively. , , Historical secondary output current data after time stabilization , , Dynamic and static characteristics; S22. The STL algorithm is used to decompose the dynamic features to obtain dynamic trend features, dynamic periodic features, and dynamic residual features: in, , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; , , They represent Dynamic trend characteristics, dynamic periodic characteristics, and dynamic residual characteristics; S3. Use the feature contribution method to determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature; The specific process of step S3 is as follows: in, As for contribution level, For the first One characteristic, For the generalized function of the LSTM algorithm, The average of all predicted feature values. It is a set that includes all dynamic and static features; S4. The optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature are fused to obtain the fused feature; S5. Input the fused features into the LSTM to obtain the error of the current transformer.

2. The current transformer error state prediction method based on feature decoupling according to claim 1, characterized in that, The stabilization process in step S1 is as follows: Where i represents the number of current transformers, and w represents the number of sampling points. This represents the average value of historical secondary output current data. This represents the variance of the historical secondary output current data. This represents the historical secondary output current data after stabilization. This represents the cross product.

3. The current transformer error state prediction method based on feature decoupling according to claim 1, characterized in that, The extraction process of dynamic and static features in step S21 is as follows: Both the dynamic and static encoders employ five layers of 4×4 convolutional operations: batch normalization and Leaky ReLU activation are applied after the first four convolutional layers, while the output layer uses the Tanh activation function to normalize the output motion feature vector to the range of -1 to 1. in, , , They represent , , Dynamic characteristics; , , They represent , , Static characteristics.

4. The current transformer error state prediction method based on feature decoupling according to claim 1, characterized in that, The specific process of step S4 is as follows: in, Let α represent the activation function, Avg represent the average pooling function, and Covn represent the convolution function, with α = 1, 2, 3, or 4. This represents the weighted weight of the α-th feature. This indicates the characteristics after fusion.

5. The current transformer error state prediction method based on feature decoupling according to claim 1, characterized in that, The specific process of step S5 is as follows: Where Y represents the error state of the current transformer, and lstm is the generalized function of the LSTM algorithm.

6. A current transformer error state prediction system based on feature decoupling, characterized in that, The method for predicting the error state of a current transformer based on feature decoupling as described in any one of claims 1-5 includes: The data acquisition and preprocessing module is used to acquire historical secondary output current data of a three-phase current transformer in a substation during operation, and to perform stabilization processing on this historical secondary output current data. The feature extraction module is used to decouple the adjacent data obtained after stabilization to extract static and dynamic features, and to decompose the dynamic features to extract dynamic trend features, dynamic periodic features and dynamic residual features. The fusion module is used to determine the optimal static feature, optimal dynamic feature, optimal dynamic trend feature, optimal dynamic periodic feature, and optimal dynamic residual feature, and to fuse these features to obtain the fused feature. The output module uses LSTM to process the fused features to obtain the error of the current transformer.

7. A storage medium, characterized in that, The system stores a program that, when executed by a processor, implements the current transformer error state prediction method based on feature decoupling as described in any one of claims 1-5.

8. A computer device, comprising a processor and a memory for storing a processor-executable program, characterized in that, When the processor executes the program stored in the memory, it implements the current transformer error state prediction method based on feature decoupling as described in any one of claims 1-5.

Images