Micro-grid fault dynamic identification method based on Gaussian mixture K-L divergence
By combining the Gaussian mixture model and KL divergence, the problems of high computational complexity and poor dynamic adaptability in microgrid fault identification are solved, achieving high-precision and real-time fault identification and meeting the rapid response requirements of microgrids.
Patent Information
- Application Number
- CN202511227734.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-29
- Publication Date
- 2026-02-10
AI Technical Summary
Existing deep learning models are computationally complex and time-consuming, making it difficult to meet the real-time monitoring needs of microgrids. Traditional methods have poor adaptability when dynamically identifying microgrid faults, lack effective probabilistic evaluation methods, and have strong uncertainty in fault signals.
A Gaussian mixture model (GMM) is used to extract features and reduce the dimensionality of microgrid fault signals. PCA is used to reduce the dimensionality, and KL divergence is used to evaluate the differences in fault states, thereby achieving dynamic identification of fault types.
It achieves high-precision, real-time fault identification with strong anti-interference capability, meets the millisecond-level response requirements of microgrids, reduces operation and maintenance costs, achieves a fault classification accuracy of 98.20%, and has a processing time of only 22.38ms.
Smart Images
Figure CN121502437A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of power system fault detection, and relates to a micro-grid fault dynamic identification method based on a mixed Gaussian K-L divergence. BACKGROUND
[0002] A micro-grid is a flexible and efficient distributed power system. It shows high flexibility and adaptability. In the current global energy transformation background, the advantages of micro-grid are more and more obvious. The stability of the operation state has a profound impact on the safety and reliability of the entire power grid. The operation mode of the micro-grid is flexible and changeable, so it is easy to be disturbed by various factors and fail during operation. These failures not only directly affect the power supply quality and reliability of the micro-grid, but also may impact the large power grid connected thereto, causing a chain reaction and threatening the stable operation of the entire power system. For example, the failure of a distributed power source or load mutation in a micro-grid may cause power fluctuation, thereby affecting the frequency and voltage stability of the large power grid; the failure of an energy storage system may cause energy supply interruption, which has a negative impact on the power balance of the power grid. Therefore, it is essential to quickly and accurately identify the faults in the micro-grid to ensure the safe and stable operation of the micro-grid and the large power grid.
[0003] Existing deep learning models (such as BiGRU and 2D-ResNet) require a large amount of training data and have a long calculation time, which makes it difficult to meet the real-time monitoring needs of the micro-grid. The fault features of the micro-grid may change over time (such as evolving from voltage sag to harmonic pollution), and traditional methods are difficult to dynamically adjust the identification strategy. Due to factors such as fluctuation of new energy output and load mutation, the fault signal often has strong uncertainty, and traditional methods lack effective probabilistic evaluation means. SUMMARY
[0004] The purpose of the present application is to provide a micro-grid fault dynamic identification method based on a mixed Gaussian K-L divergence. First, the time-frequency domain multi-dimensional feature parameters of the micro-grid voltage signal are extracted, and PCA is used for dimension reduction to reduce the feature dimension. A mixed Gaussian model (GMM) is established to represent the fault feature distribution, and K-L divergence is used to evaluate the difference between fault states to realize dynamic identification of fault types. The present application solves the problems of high computational complexity and poor dynamic adaptability of traditional methods, has high precision, high real-time performance and strong anti-interference ability, and is suitable for micro-grid fault identification.
[0005] TECHNICAL SOLUTION: The micro-grid fault dynamic identification method based on a mixed Gaussian K-L divergence of the present application comprises the following steps:
[0006] Step 1, feature parameter extraction is performed on the fault signal;
[0007] Step 2, PCA dimension reduction is performed on the original fault features;
[0008] Step 3: Input the dimensionality-reduced features into the GMM algorithm to perform multidimensional feature mixture Gaussian analysis;
[0009] Step 4: Use KL divergence to assess the monitoring distribution and identify the fault type.
[0010] Furthermore, step 1 specifically includes:
[0011] Let b(t) be the voltage signal of the microgrid under normal conditions, i.e., the reference signal. m(t) is the real-time acquired operating voltage signal. Parameters are extracted from the six voltage signals, i.e.
[0012] (1) DI1 characterizes the correlation between the reference signal and the monitoring signal.
[0013]
[0014] Where t1 and t2 are the start and end times of the valid signal, respectively.
[0015] (2) DI2 is the root mean square of the time-domain signal.
[0016]
[0017] (3) DI3 is the frequency response difference between the real-time signal and the reference signal.
[0018]
[0019] Where ω0 and ω1 are the start and end frequencies of the effective signal spectrum amplitude, respectively, and F b (ω), F m (ω) represents the frequency response of b(t) and m(t), respectively.
[0020] (4) DI4 is the difference in frequency response amplitude between the real-time signal and the reference signal. This parameter is used to measure the energy dependence of signal changes.
[0021]
[0022] (5) DI5 is the normalized correlation moment, used to evaluate the phase and amplitude variation characteristics.
[0023]
[0024] Where r bm (τ) is the cross-correlation coefficient between the reference signal and the monitoring signal, r bb (τ) is the cross-correlation coefficient between the reference signal and itself. n is the order of the statistical moments.
[0025] (6) DI6 is the relative area of the time-domain signal envelope, representing the relative energy of the reference signal and the monitoring signal.
[0026]
[0027] Furthermore, step 2 specifically includes:
[0028] Step 201, let the feature set of microgrid fault signal parameters be D = (DI1, DI2, ... DI n Standardize the feature parameters:
[0029]
[0030] Among them, DI′ i It is the standardized feature parameter, μ i It is the characteristic parameter DI i The mean, σ i It is its standard deviation.
[0031] Step 202: Calculate the covariance matrix for the standardized data. This matrix measures the correlation between different features. Assuming the standardized data matrix is X with dimensions m×n (m is the number of samples, n is the number of features), the formula for calculating the covariance matrix Σ is:
[0032]
[0033] Among them, X T It is the transpose of X.
[0034] Step 203: Perform eigenvalue decomposition on the covariance matrix Σ to solve for its eigenvalues and eigenvectors. Eigenvalues represent the magnitude of the variance of each principal component, while eigenvectors represent the direction of the principal components. Sort the eigenvalues from largest to smallest, and select the eigenvectors corresponding to the top k eigenvalues to form the dimensionality reduction matrix W. The directions corresponding to these eigenvectors are those with the largest variance in the data, thus preserving the most original data information.
[0035] Step 204: Use the dimensionality reduction matrix W to reduce the dimensionality of the original data:
[0036] X PCA =X*W
[0037] Among them, X PCA This is the data after PCA dimensionality reduction, with a dimension of m×k.
[0038] Furthermore, step 3 specifically includes:
[0039] The reduced feature set is D = (DI1, DI2, ..., DI) kThe probability density function (PDF) of a Gaussian mixture distribution (GMM) is:
[0040]
[0041] Where P is the number of Gaussian components, ω i μ represents the weight of each component. i Let ∑ be the mean of each component. i Let N be the covariance. i The PDF for each Gaussian component is:
[0042]
[0043] Where Q represents the distribution dimension.
[0044] Furthermore, step 4 specifically includes:
[0045]
[0046] Among them, KL(N) j, N i ) is N i and N j The KL divergence. Tr is the trace of the matrix, det is the determinant of the matrix, De is the sample dimension, and N is the number of samples. i The fault distribution is Gaussian, N j As the baseline distribution, μ i and μ j These are the mean vectors of the fault Gaussian distribution and the baseline distribution, respectively. ∑ i and ∑ j These are the covariance matrices of the fault Gaussian distribution and the baseline distribution, respectively. Fault identification is achieved by evaluating the differences between the fault Gaussian distribution and the baseline distribution.
[0047] This invention also proposes a microgrid fault dynamic identification system based on mixed Gaussian KL divergence, which uses the aforementioned microgrid fault dynamic identification method based on mixed Gaussian KL divergence to perform microgrid fault dynamic identification.
[0048] The present invention also proposes a computer storage medium having computer instructions, which, when invoked, are used to execute the microgrid fault dynamic identification method based on mixed Gaussian KL divergence.
[0049] Compared with existing technologies, the beneficial effects are:
[0050] 1. The time-frequency domain multidimensional features (DI1~DI6) of microgrid faults are probabilistically modeled using a Gaussian mixture model (GMM), replacing the traditional fixed threshold method. The fault classification accuracy reaches 98.20% (better than BiGRU's 94.57% and 2D-ResNet's 96.68%).
[0051] 2. By employing PCA to reduce feature dimensionality by over 50% and combining it with lightweight GMM modeling, the average processing time is only 22.38ms, far lower than 2D-ResNet (143.91ms) and LSTM-Attention (77.52ms), meeting the millisecond-level fault response requirements of microgrids. Furthermore, it does not rely on a large amount of labeled data; only a small amount of baseline data is needed to build the GMM model, reducing operation and maintenance costs by 40%. Attached Figure Description
[0052] Figure 1 The flowchart below shows a method for dynamic identification of microgrid faults based on mixed Gaussian KL divergence according to the present invention.
[0053] Figure 2 This is a simulation system topology diagram of the present invention;
[0054] Figure 3 This is a simulation signal diagram of the present invention;
[0055] Figure 4 This is a diagram of the characteristic parameters of the present invention;
[0056] Figure 5 This is the GMM distribution map of the present invention;
[0057] Figure 6 This is a diagram showing the differences in the results of this invention. Detailed Implementation
[0058] The following detailed description, in conjunction with the accompanying drawings, illustrates a specific implementation method for dynamic fault identification in microgrids based on mixed Gaussian KL divergence according to the present invention.
[0059] like Figure 1 As shown, the present invention provides an intelligent diagnosis method for complex faults in power distribution network transmission lines, the method comprising the following steps:
[0060] Step 1: Extract feature parameters from the fault signal;
[0061] Step 2: Perform PCA dimensionality reduction on the original fault features;
[0062] Step 3: Input the dimensionality-reduced features into the GMM algorithm to perform multidimensional feature mixture Gaussian analysis;
[0063] Step 4: Use KL divergence to assess the monitoring distribution and identify the fault type.
[0064] A Digital Dynamic Real-Time Simulator (DDRTS) system is used to simulate the microgrid operating environment and generate faults. The topology diagram of the DDRTS system is shown below. Figure 2 As shown. The six types of fault signals obtained from the simulation are as follows: Figure 3 As shown.
[0065] Step 1 specifically includes:
[0066] Let b(t) be the voltage signal of the microgrid under normal conditions, i.e., the reference signal. m(t) is the real-time acquired operating voltage signal. Parameters are extracted from the six voltage signals, i.e.
[0067] (1) DI1 characterizes the correlation between the reference signal and the monitoring signal.
[0068]
[0069] Where t1 and t2 are the start and end times of the valid signal, respectively.
[0070] (2) DI2 is the root mean square of the time-domain signal.
[0071]
[0072] (3) DI3 is the frequency response difference between the real-time signal and the reference signal.
[0073]
[0074] Where ω0 and ω1 are the start and end frequencies of the effective signal spectrum amplitude, respectively, and F b (ω), F m (ω) represents the frequency response of b(t) and m(t), respectively.
[0075] (4) DI4 is the difference in frequency response amplitude between the real-time signal and the reference signal. This parameter is used to measure the energy dependence of signal changes.
[0076]
[0077] (5) DI5 is the normalized correlation moment, used to evaluate the phase and amplitude variation characteristics.
[0078]
[0079] Where r bm (τ) is the cross-correlation coefficient between the reference signal and the monitoring signal, r bb(τ) is the cross-correlation coefficient between the reference signal and itself. n is the order of the statistical moments.
[0080] (6) DI6 is the relative area of the time-domain signal envelope, representing the relative energy of the reference signal and the monitoring signal.
[0081]
[0082] Six characteristic parameters of different fault data, such as Figure 4 As shown.
[0083] Step 2 specifically includes:
[0084] Step 201, let the feature set of microgrid fault signal parameters be D = (DI1, DI2, ... DI n Standardize the feature parameters:
[0085]
[0086] Among them, DI′ i It is the standardized feature parameter, μ i It is the characteristic parameter DI i The mean, σ i It is its standard deviation.
[0087] Step 202: Calculate the covariance matrix for the standardized data. This matrix measures the correlation between different features. Assuming the standardized data matrix is X with dimensions m×n (m is the number of samples, n is the number of features), the formula for calculating the covariance matrix Σ is:
[0088]
[0089] Among them, X T It is the transpose of X.
[0090] Step 203: Perform eigenvalue decomposition on the covariance matrix Σ to solve for its eigenvalues and eigenvectors. Eigenvalues represent the magnitude of the variance of each principal component, while eigenvectors represent the direction of the principal components. Sort the eigenvalues from largest to smallest, and select the eigenvectors corresponding to the top k eigenvalues to form the dimensionality reduction matrix W. The directions corresponding to these eigenvectors are those with the largest variance in the data, thus preserving the most original data information.
[0091] Step 204: Use the dimensionality reduction matrix W to reduce the dimensionality of the original data:
[0092] X PCA =X*W
[0093] Among them, X PCA This is the data after PCA dimensionality reduction, with a dimension of m×k.
[0094] Step 3 specifically includes:
[0095] The reduced parameter feature set is D = (DI1, Di2, ..., DI) k The probability density function (PDF) of a Gaussian mixture distribution (GMM) is:
[0096]
[0097] Where P is the number of Gaussian components, ω i μ represents the weight of each component. i Let ∑ be the mean of each component. i Let N be the covariance. i The PDF for each Gaussian component is:
[0098]
[0099] Where Q represents the distribution dimension. GMM distributions with different parameters are as follows: Figure 5 As shown.
[0100] Step 4 specifically includes:
[0101]
[0102] Among them, KL(N) j, N i ) is N i and N j The KL divergence. Tr is the trace of the matrix, det is the determinant of the matrix, De is the sample dimension, and N is the number of samples. i The fault distribution is Gaussian, N j As the baseline distribution, μ i and μ j These are the mean vectors of the fault Gaussian distribution and the baseline distribution, respectively. i and ∑ j These are the covariance matrices of the fault's Gaussian distribution and the baseline distribution, respectively. Fault identification is achieved by evaluating the differences between the fault's Gaussian distribution and the baseline distribution. The results of the differences are as follows: Figure 6 As shown in the table, the membership probability is determined based on the differences in the results. Table 1 shows a comparison of the running time of the algorithm presented in this paper with that of other algorithms. Table 2 shows a comparison of the evaluation metrics of the algorithm presented in this paper with that of other algorithms.
[0103] Table 1
[0104]
[0105]
[0106] Table 2
[0107]
[0108] The present invention also proposes a microgrid fault dynamic identification system based on mixed Gaussian KL divergence, which uses a microgrid fault dynamic identification method based on mixed Gaussian KL divergence to perform microgrid fault dynamic identification.
[0109] The present invention also proposes a computer storage medium having computer instructions, which, when invoked, are used to execute a microgrid fault dynamic identification method based on mixed Gaussian KL divergence.
[0110] Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and not intended to limit it. Those skilled in the art should understand that modifications or equivalent substitutions can be made to the specific embodiments of the present invention, but such modifications or alterations are all within the scope of protection of the pending claims.
Claims
1. A dynamic fault identification method for microgrids based on mixed Gaussian KL divergence, characterized in that, The microgrid fault dynamic identification method includes: Step 1: Extract feature parameters from the fault signal; Step 2: Perform PCA dimensionality reduction on the original fault features; Step 3: Input the dimensionality-reduced features into the GMM algorithm to perform multidimensional feature mixture Gaussian analysis; Step 4: Use KL divergence to assess the monitoring distribution and identify the fault type.
2. The microgrid fault dynamic identification method based on mixed Gaussian KL divergence according to claim 1, characterized in that, The specific method for extracting feature parameters from the fault signal in step 1 is as follows: Let b(t) be the voltage signal of the microgrid under normal conditions, i.e., the reference signal; m(t) be the real-time acquired operating voltage signal; extract the parameters of the six voltage signals, i.e. (1) DI1 characterizes the correlation between the reference signal and the monitoring signal. Where t1 and t2 are the start and end times of the valid signal, respectively; (2) DI2 is the root mean square of the time-domain signal. (3) DI3 is the frequency response difference between the real-time signal and the reference signal. Where ω0 and ω1 are the start and end frequencies of the effective signal spectrum amplitude, respectively, and F b (ω), F m (ω) represent the frequency responses of b(t) and m(t), respectively; (4) DI4 is the frequency response amplitude difference between the real-time signal and the reference signal. This parameter is used to measure the energy dependence of signal changes. (5) DI5 is the normalized correlation moment, used to evaluate the phase and amplitude variation characteristics. Where r bm (τ) is the cross-correlation coefficient between the reference signal and the monitoring signal, r bb (τ) is the cross-correlation coefficient between the reference signal and itself; n is the order of the statistical moments; (6) DI6 is the relative area of the time-domain signal envelope, representing the relative energy of the reference signal and the monitoring signal.
3. The microgrid fault dynamic identification method based on mixed Gaussian KL divergence according to claim 2, characterized in that, Step 2 involves performing PCA dimensionality reduction on the original fault features. The specific steps are as follows: Step 201, let the feature set of microgrid fault signal parameters be D = (DI1, DI2, ... DI n Standardize the feature parameters: Among them, DI′ i It is the standardized feature parameter, μ i It is the characteristic parameter DI i The mean, σ i It is its standard deviation; Step 202: Calculate the covariance matrix for the standardized data; this is used to measure the correlation between different features. Assuming the standardized data matrix is X, with dimensions m×n, where m is the number of samples and n is the number of features, the formula for calculating the covariance matrix Σ is: Among them, X T It is the transpose of X; Step 203: Perform eigenvalue decomposition on the covariance matrix Σ to solve for its eigenvalues and eigenvectors; the eigenvalues represent the variance of each principal component, and the eigenvectors represent the direction of the principal components; sort the eigenvalues from largest to smallest, select the eigenvectors corresponding to the first k eigenvalues to form the dimensionality reduction matrix W; the directions corresponding to these eigenvectors are the directions with the largest variance in the data, which can retain the most original data information; Step 204: Use the dimensionality reduction matrix W to reduce the dimensionality of the original data: X PCA =X*W Among them, X PCA This is the data after PCA dimensionality reduction, with a dimension of m×k.
4. The microgrid fault dynamic identification method based on mixed Gaussian KL divergence according to claim 3, characterized in that, The specific steps for performing multidimensional feature mixture Gaussian analysis in step 3 are as follows: The reduced feature set is D = (DI1, DI2, ..., DI) k The probability density function (PDF) of a Gaussian mixture distribution (GMM) is: Where P is the number of Gaussian components, ω i μ represents the weight of each component. i Let ∑ be the mean of each component. i For covariance; N i The PDF for each Gaussian component is: Where Q represents the distribution dimension.
5. The microgrid fault dynamic identification method based on mixed Gaussian KL divergence according to claim 4, characterized in that, In step 4, fault type identification is achieved by using KL divergence to assess the monitoring distribution. The specific steps are as follows: Among them, KL(N) j, N i ) is N i and N j KL divergence; Tr is the trace of the matrix, det is the determinant of the matrix, De is the sample dimension, and N is the number of samples. i The fault distribution is Gaussian, N j As the baseline distribution, μ i and μ j These are the mean vectors of the fault Gaussian distribution and the baseline distribution, respectively; ∑ j and ∑ i These are the covariance matrices of the fault Gaussian distribution and the baseline distribution, respectively; fault identification is achieved by evaluating the differences between the fault Gaussian distribution and the baseline distribution.
6. A microgrid fault dynamic identification method system based on mixed Gaussian KL divergence, characterized in that, The microgrid fault dynamic identification method based on mixed Gaussian KL divergence as described in any one of claims 1-5 is used for microgrid fault dynamic identification.
7. A computer storage medium, characterized in that, The computer storage medium contains computer instructions, which, when invoked, are used to execute the microgrid fault dynamic identification method based on mixed Gaussian KL divergence as described in any one of claims 1-5.