Voltage transformer on-line monitoring method based on independent component analysis

Abstract

The invention discloses a voltage transformer on-line monitoring method based on independent component analysis, and the method comprises the steps: collecting historical data, steady-state data and real-time data of an output signal of a voltage transformer, carrying out the information sampling, and constructing a data set through the historical data; importing the data set into an initial SDAE network model constructed on the basis of a sparse denoising automatic encoder for dynamic training, and performing parameter fine tuning by using steady-state data on the basis of an SDAE network model obtained by offline training so as to obtain coded data of the SDAE network model; taking the coded data as input, and performing independent component decomposition by using an independent component analysis method; calculating a sample statistical magnitude and an overall statistical threshold value, comparing the real-time statistical magnitude with the overall statistical threshold value, and if the real-time statistical magnitude is lower than the overall statistical threshold value, determining that the state of the mutual inductor in the judgment is normal; otherwise, it is considered that the mutual inductor in the mutual inductor group may be abnormal.

Description

Technical Field

[0001] The present invention relates to the technical field of online monitoring of power equipment, and in particular to an online monitoring method for a voltage transformer based on independent component analysis. Background Art

[0002] In modern power systems, voltage transformers, as a crucial electrical device, play an indispensable role. They are mainly used to convert high voltage into low voltage proportionally, providing accurate and reliable voltage signals for measurement, protection, and control of power systems. The accuracy of their measurements and the reliability of their operation are directly related to the safe and stable operation of power systems, power quality assessment, and the accuracy of power metering.

[0003] During actual power system operation, voltage transformers face complex problems such as three-phase imbalance and harmonic dynamic disturbances. On the one hand, three-phase voltage imbalance is common due to factors such as asymmetric three-phase loads and the connection of single-phase large-capacity loads. Three-phase imbalance can cause imbalance in the three-phase excitation current of the voltage transformer, thereby increasing the transformer's error and affecting measurement accuracy. If the transformer is in a three-phase unbalanced operating environment for a long time, it may also accelerate its insulation aging and shorten its service life. On the other hand, with the widespread application of power electronics technology, a large number of nonlinear loads (such as inverters, rectifiers, arc furnaces, etc.) are connected to the power system, generating a wealth of harmonics. Harmonics can saturate the core of the voltage transformer, resulting in increased excitation current and losses, while affecting the frequency response characteristics of the transformer and causing large errors in the measurement results. Moreover, the dynamic change characteristics of harmonics make the operating status of the transformer more complex and greatly increase the difficulty of monitoring and diagnosis. Therefore, based on the above technical problems, this application proposes an online monitoring method for voltage transformers based on independent component analysis, which can effectively process data characteristics, improve monitoring accuracy, and enhance model robustness. Summary of the Invention

[0004] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a voltage transformer online monitoring method based on independent component analysis that can effectively process data characteristics, improve monitoring accuracy, and enhance model robustness.

[0005] In order to achieve the above-mentioned object, the present invention provides a voltage transformer online monitoring method based on independent component analysis, which is characterized by comprising the following steps: S1 collects historical data, steady-state data and real-time data based on the output signal of the substation voltage transformer group, and constructs a data set based on the historical data; S2 creates the initial SDAE network model through the sparse denoising autoencoder; S3 imports the dataset into the initial SDAE network model created in step S2 for offline training, thereby obtaining the optimal SDAE network model; S4 imports the steady-state data into the SDAE network model for correction training. When the SDAE network model training times reaches a preset value or the iteration effect reaches a preset purpose, a corrected SDAE network model is obtained. S5 imports the steady-state data into the SDAE network model for encoding and then performs data standardization; S6 takes the standardized coded data as input, performs independent component decomposition through independent component analysis, and calculates the sample statistics. ; S7 calculates sample statistics using kernel density estimation The overall statistical threshold ; S8 collects raw data in real time and calculates real-time statistics through steps S5-S6 ; S9 will provide real-time statistics and overall statistical threshold By comparison, it is determined whether the voltage transformer group has a fault. Greater than the overall statistical threshold When the real-time statistics Less than the overall statistical threshold When , the voltage transformer group has no fault; S10 repeats steps S1-S9 to perform real-time evaluation on the voltage transformer group.

[0006] Furthermore, we introduce the data obtained by mixing a certain proportion of noise signal with the input sample. , used to represent the weight coefficient for controlling the sparse penalty term , represents the number of neurons in the hidden layer , used to represent the sparsity parameter And it is used to represent the average activation of the jth neuron under the sample data , thereby constructing the loss function of the SDAE network model.

[0007] Furthermore, the independent component analysis method ICA performs independent component decomposition including the following steps: A1 steady-state data matrix construction, decomposing the steady-state data into a mixing matrix A and an independent component matrix S; A2 data standardization, standardize the steady-state data matrix to obtain the matrix , so that the mean of each column element is 0 and the standard deviation is 1; A3 data whitening, which eliminates data covariance and achieves independence and the same variance between vectors; A4 solves the problem conversion, thereby converting the solution of the demixing matrix W into the solution of the transformation matrix B; A5 solves the transformation matrix B to extract independent components; A6 independent component decomposition, decomposing the unmixing matrix W, the mixing matrix A and the independent component S; A7 component separation, which divides the unmixing matrix W and the transformation matrix B into main components and residual components; A8 Calculate sample statistics ; A9 uses kernel density estimation to calculate the overall statistical threshold of the corresponding confidence level .

[0008] The present invention adopts the above-mentioned solution, and its beneficial effects are: 1) Effectively handle data characteristics and improve monitoring accuracy: By integrating the SDAE network model and independent component analysis method, it effectively addresses the nonlinear and interference characteristics of transformer monitoring data, greatly improving the accuracy of voltage transformer operating status monitoring; 2) Enhanced model robustness and adaptability to complex environments: During SDAE network model training, random noise is added to the input data, and a sparse penalty term is introduced to force the model to learn relatively sparse and concise numerical features. This reduces the risk of overfitting, enabling the model to better adapt to complex and changing operating environments, thereby enhancing the model's resistance to noise and interference. 3) Dynamic model optimization to ensure monitoring reliability: By adopting incremental learning, the model obtained through offline training is fine-tuned using real-time data collected online to adapt to dynamic changes in data in a timely manner and continuously optimize the model's own performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 A schematic flow chart of a method provided by an embodiment of the present invention.

[0010] Figure 2 Schematic diagram of the process of the independent component analysis method in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] To facilitate understanding of the present invention, the present invention is described more fully below with reference to the accompanying drawings. The accompanying drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein. These embodiments are provided solely to provide a more thorough and comprehensive understanding of the present disclosure.

[0012] See attached Figure 1 As shown, in this embodiment, a voltage transformer online monitoring method based on independent component analysis is characterized by comprising the following steps: S1 collects historical data, steady-state data, and real-time data based on the output signals of the substation voltage transformer group, and constructs a data set through historical data, so as to provide data support for the subsequent training of the SDAE network model and improve the accuracy of model prediction; specifically, historical data refers to the output signals of the voltage transformer group collected in the past and the output signals of related voltage transformer groups collected by other substations; steady-state data refers to data from the first week of grid operation (this type of data is steady-state by default), which is used to generate initial output weights and initial matrices; real-time data refers to the real-time output signals of the voltage transformer group collected immediately.

[0013] S2 uses sparse denoising autoencoder SDAE to create the initial SDAE network model; Specifically, the above-mentioned autoencoder (AE) is a kind of neural network, including an input layer, a hidden layer and an output layer, wherein the input layer is used for data input, the hidden layer represents the learned feature expression, and the output layer represents the reconstructed value after the decoding process; the autoencoder is an unsupervised learning algorithm, whose purpose is to let the output value try to reconstruct the input value rather than obtain the target value of the input x after training. Secondly, the autoencoder network consists of two parts, one is composed of a function An encoder that represents the representation and a decoder that generates the reconstruction In order to make the output reconstruct the input as much as possible, the autoencoder must extract the most important features representing the input, which usually requires that the number of nodes in the hidden layer of the autoencoder is smaller than the number of nodes in the input layer; During the encoding process, the nonlinear activation function Map the input data x to the hidden layer h, the mapping formula is: , , in yes weight matrix, is the bias vector; During decoding, the activation function Map the hidden layer expression h to the input layer , the mapping formula is: , in, Generally speaking and The function is consistent with the sigmoid activation function; yes The weight matrix of this embodiment is preferably set to , also commonly called binding weight; is the bias vector, The sparse denoising autoencoder (SDAE) used in this embodiment combines the advantages of the sparse autoencoder (SAE) and the denoising autoencoder (DAE). First, a sparse penalty term is introduced on the basis of the autoencoder loss function, and relatively sparse and concise digital features are learned under the sparsity constraint, so that the characteristic distribution of the input data can be effectively obtained. At the same time, random noise is superimposed on the input data to enhance the robustness of the process monitoring model and obtain strong generalization ability. The above method uses comprehensive statistical analysis information to improve the accuracy and robustness of the process monitoring model in view of the nonlinear and interference characteristics of the mutual inductor monitoring data. Furthermore, relevant parameters are introduced to construct the loss function of the SDAE network model: , The loss function of the autoencoder is the sum of the input data x and the reconstructed data The sum of the reconstruction errors between , that is, to minimize the reconstruction error as the goal to find The optimal parameters of Where L is the reconstruction error loss function, which is generally the sum of squared errors (the loss value L reflects the performance of the model under the current model parameters); n is the number of samples; represents the i-th sample in the input data, It represents the data obtained after the input sample is mixed with a certain proportion of noise signal. represents the weight coefficient controlling the sparse penalty term, represents the number of neurons in the hidden layer, represents the sparsity parameter, represents the average activation of the jth neuron under the sample data, where Divergence describes the difference between two probability distributions, and its value varies with the two probability distributions. and As the distance between them increases, it increases monotonically. hour, Get the minimum value; in, and The specific expression is as follows: , , in, Represents the activation value of the jth neuron in the autoencoder.

[0014] S3 imports the data set into the initial SDAE network model created in step S2 for offline training to obtain the optimal SDAE network model. By creating the initial SDAE network model based on historical data, a relatively large number of data samples can be provided for related model establishment, thereby improving the accuracy of the model's operation, reducing the time required for subsequent model corrections, and relatively reducing investment costs.

[0015] S4 imports the steady-state data into the SDAE network model for correction training. When the number of SDAE network model training times reaches the preset value or the iterative effect reaches the preset purpose, the corrected SDAE network model is obtained. Specifically, the preset value of the first batch of training times of the SDAE network model or the preset purpose of the iterative effect can be set according to actual conditions. No specific restrictions are made here. Further correction of the SDAE network model through steady-state data can reduce the calculation error of the SDAE network model and improve its calculation accuracy. Secondly, steady-state data, that is, the relatively constant output signal of the voltage transformer over a long period of time, is more suitable for the subsequent SDAE network model to perform overall statistical thresholding. Calculation speculation.

[0016] S5 imports the steady-state data into the SDAE network model for encoding and then performs data standardization.

[0017] S6 takes the standardized coded data as input and performs independent component decomposition using the independent component analysis method (ICA method). Figure 2 As shown, calculate the sample statistics ; Specifically, the independent component analysis method includes the following steps to perform independent component decomposition: A1 steady-state data matrix construction, which is constructed by collecting steady-state data , where n is the number of samples and m is the number of elements in a single sample acquisition; thus, the following data model is established to decompose the original data into a mixing matrix A and an independent component matrix S (n×r, where r is the number of independent components), namely: ; Since the column vectors (i.e., independent components) of the independent component matrix S satisfy statistical independence and have zero mean and unit variance characteristics, the problem can be converted to: the sample matrix X is transformed by the unmixing matrix W, and the resulting matrix S has statistical independence characteristics, that is: ; A2 data standardization, standardize the steady-state data matrix to obtain the matrix , so that the mean of each column element is 0 and the standard deviation is 1, and it can also replace X in the mathematical model in the subsequent solution, that is, the solution problem is converted to: ; A3 Data whitening: Data covariance is eliminated through data whitening to achieve mutual independence (no correlation between vectors) and the same variance between vectors. The specific process of data whitening is as follows: First solve the covariance matrix of the observation matrix : ; Pair Matrix Perform eigenvalue decomposition to obtain its eigenvector matrix and the diagonal matrix consisting of eigenvalues : ; Solving the whitening matrix : ; Solve the matrix Z after data whitening: ; A4 solves the problem transformation. Matrix B is a transformation of the mixing matrix A. Compared with the full-rank mixing matrix A, matrix B is orthogonal, so there are fewer parameters to be estimated. The specific derivation is as follows: , According to the above derivation, it can be deduced that the independent component matrix S and the whitened matrix Z satisfy: , Further convert the solution of the demixing matrix W into the solution of the conversion matrix B: , A5 solves the transformation matrix B. The independent component decomposition method separates independent source signals through the maximum independence criterion, and independence can be achieved by minimizing mutual information or maximizing non-Gaussianity. Since the Gaussian distribution has the largest entropy, non-Gaussianity can be used as an indicator of independence. By maximizing non-Gaussianity, independent components can be extracted. The independent component analysis method uses the above idea to implement the following iterative process to solve the matrix B: ①Determine the number of independent components r that need to be estimated, and remember i = 1; ② Randomly select the initial unit vector (column vector of B); ③Use the following formula to Perform amplitude calculation, where E{} represents expectation; , in, express The derivative of is a nonlinear function, which can be selected from the following functions: , , , ④ On income Perform standard orthogonalization processing; ⑤ If the If it does not converge, return to step ③; ⑥ If you get Convergence, output vector ; ⑦ If i < r, set i = i + 1 and return to step ②. If i = r, end the above process. ⑧ All output column vectors Construct matrix B.

[0018] A6 independent component decomposition decomposes the unmixing matrix W, the mixing matrix A and the independent component S. The specific calculation method is: , , .

[0019] A7 component separation, the unmixing matrix W and the conversion matrix B are divided into the main component and the residual component. Specifically, the L2 norm of the column vector of the mixing matrix A is calculated. (i = 1, 2, ..., r, r is the number of independent components); Secondly, arrange the required L2 norms in descending order and calculate the contribution of the feature quantity and cumulative contribution ; , ;

[0020] For example, it is preferred to use the cumulative contribution exceeding 85% as the standard to divide all components into main components (column vectors corresponding to feature quantities that participate in 85% of the cumulative contribution) and residual components; The unmixing matrix W is divided into the main part according to the row vector (the row index mapped by the column index of A) and the remaining ; Divide the transformation matrix B into the main part according to the column vector (the column index corresponding to the column index of A) and the remaining .

[0021] A8 Calculate sample statistics , in calculating the sample statistics (sample squared prediction error), calculate the i-th sample vector The formula for the statistic is: , in, Represents the reconstructed variables of the main components separated by the independent component analysis method at this moment: , According to the formula Will The calculation is further simplified to: , S7 calculates sample statistics using kernel density estimation The overall statistical threshold ; A9 uses kernel density estimation to calculate the overall statistical threshold of the corresponding confidence level , where the above confidence level refers to the degree to which a specific individual believes in the truth of a specific proposition, that is, probability is a measure of the rationality of an individual's belief; S8 collects raw data in real time and calculates real-time statistics through steps S5-S6 ; S9 will provide real-time statistics and overall statistical threshold By comparison, it is determined whether the voltage transformer group has a fault. Greater than the overall statistical threshold When the real-time statistics Less than the overall statistical threshold When , the voltage transformer group has no fault; S10 repeats steps S1-S9 to perform real-time evaluation on the voltage transformer group.

[0022] Through continuous training and real-time evaluation and comparison in the above steps, the SDAE network model used can be trained as data is imported, thereby improving the accuracy and robustness of the process monitoring model. In practical applications, it is also necessary to combine multiple strategies such as superimposed random noise processing of updated data, model architecture selection, and related parameter tuning to further improve the performance of the model.

[0023] In summary, the above advantages include: 1) Effectively process data characteristics and improve monitoring accuracy: By integrating the SDAE network model and independent component analysis method, the nonlinear and interference characteristics of transformer monitoring data are effectively addressed, greatly improving the accuracy of voltage transformer operating status monitoring; 2) Enhanced model robustness and adaptability to complex environments: During SDAE network model training, random noise is added to the input data, and a sparse penalty term is introduced to force the model to learn relatively sparse and concise numerical features. This reduces the risk of overfitting, enabling the model to better adapt to complex and changing operating environments, thereby enhancing the model's resistance to noise and interference. 3) Dynamic model optimization to ensure monitoring reliability: By adopting incremental learning, the model obtained through offline training is fine-tuned using real-time data collected online to adapt to dynamic changes in data in a timely manner and continuously optimize the model's own performance.

[0024] This embodiment takes into account the relatively stable three-phase voltage imbalance of the transformer and simultaneously addresses nonlinear problems that may be caused by three-phase imbalance or harmonic dynamic disturbances. The corresponding data is analyzed based on the Sparse Denoising Autoencoder-Independent Component Analysis (SDAE-ICA) method. The SDAE method is an improvement on the traditional encoder method and can be effectively used for data feature extraction and dimensionality reduction in nonlinear processes. The independent component analysis method (ICA method) can extract non-Gaussian components in the data and enhance the separability of fault characteristics.

[0025] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any form. Any person skilled in the art who, without departing from the scope of the technical solution of the present invention, utilizes the technical content disclosed above to make more possible changes and modifications to the technical solution of the present invention, or modifications are all equivalent embodiments of the present invention. Therefore, any equivalent and equivalent changes made in accordance with the ideas of the present invention without departing from the content of the technical solution of the present invention should be included in the scope of protection of the present invention.

Claims

1. A voltage transformer online monitoring method based on independent component analysis, characterized by: The following steps are involved: S1 collects historical data, steady-state data and real-time data based on the output signal of the substation voltage transformer group, and constructs a data set based on the historical data; S2 creates the initial SDAE network model through the sparse denoising autoencoder; S3 imports the dataset into the initial SDAE network model created in step S2 for offline training, thereby obtaining the optimal SDAE network model; S4 imports the steady-state data into the SDAE network model for correction training. When the number of SDAE network model training times reaches a preset value or the iteration effect reaches a preset purpose, a corrected SDAE network model is obtained; . S5 imports the steady-state data into the SDAE network model for encoding and then performs data standardization; S6 takes the standardized coded data as input, performs independent component decomposition through independent component analysis, and calculates the sample statistics. ; S7 calculates sample statistics using kernel density estimation The overall statistical threshold ; S8 collects raw data in real time and calculates real-time statistics through steps S5-S6 ; S9 will provide real-time statistics and overall statistical threshold By comparison, it is determined whether the voltage transformer group has a fault. Greater than the overall statistical threshold When , the voltage transformer group fails; On the contrary, when real-time statistics Less than the overall statistical threshold When , the voltage transformer group has no fault; S10 repeats steps S1-S9 to perform real-time evaluation on the voltage transformer group.

2. The method for online monitoring of a voltage transformer based on independent component analysis according to claim 1, characterized in that: Introduced to represent the data obtained after the input sample is mixed with a certain proportion of noise signal , used to represent the weight coefficient for controlling the sparse penalty term , represents the number of neurons in the hidden layer , used to represent the sparsity parameter And it is used to represent the average activation of the jth neuron under the sample data , thereby constructing the loss function of the SDAE network model.

3. The method for online monitoring of a voltage transformer based on independent component analysis according to claim 1, characterized in that: The independent component analysis method ICA performs independent component decomposition, which includes the following steps: A1 steady-state data matrix construction, decomposing the steady-state data into a mixing matrix A and an independent component matrix S; A2 data standardization, standardize the steady-state data matrix to obtain the matrix , so that the mean of each column element is 0 and the standard deviation is 1; A3 data whitening, which eliminates data covariance and achieves independence and the same variance between vectors; A4 solves the problem conversion, thereby converting the solution of the demixing matrix W into the solution of the transformation matrix B; A5 solves the transformation matrix B to extract independent components; A6 independent component decomposition, decomposing the unmixing matrix W, the mixing matrix A and the independent component S; A7 component separation, which divides the unmixing matrix W and the transformation matrix B into main components and residual components; A8 Calculate sample statistics ; A9 uses kernel density estimation to calculate the overall statistical threshold of the corresponding confidence level .

Images