Power distribution network intelligent monitoring method and system based on data medium station and edge calculation

Abstract

The invention relates to a power distribution network intelligent monitoring method based on a data middle station and edge calculation, and the method comprises the following steps: S1, arranging a sensor in a power distribution network, collecting key data in real time, and carrying out the denoising processing and normalization processing of the collected data; s2, performing preliminary anomaly detection on the processed data at the edge end by using a lightweight GMM model; s3, storing the preprocessed data from the edge nodes in the data center set, and performing secondary detection by adopting a variational auto-encoder based on a preliminary anomaly detection result to obtain a final detection result; s4, constructing a causal relationship graph of the power distribution network, and performing anomaly analysis based on a final detection result; and S5, generating a maintenance report according to the final detection result and the abnormal analysis result. According to the invention, by integrating the edge calculation and the data center, and through multi-stage detection and comprehensive analysis, the abnormity in the network is found and predicted in time, and a reliable guarantee is provided for intelligent management and optimization of the power distribution network.

Description

Technical Field

[0001] This invention relates to the field of power distribution network detection, and in particular to a method and system for intelligent monitoring of power distribution networks based on a data middleware platform and edge computing. Background Technology

[0002] As the scale and complexity of power distribution networks continue to increase, their operational stability and reliability become particularly important. Anomaly monitoring in power distribution networks, as a key component in ensuring the normal operation and fault handling of power systems, is receiving increasing attention and research. Anomaly monitoring technology aims to quickly detect, diagnose, and respond to various anomalies that may occur in power distribution networks, such as equipment failures, line overloads, short-circuit faults, voltage dips, and harmonic distortion.

[0003] Traditional power distribution network monitoring typically relies on monitoring equipment at a few key points, making it difficult to achieve real-time monitoring across the entire network and prone to blind spots. Furthermore, data acquisition and transmission speeds are slow, hindering timely responses to rapidly changing anomalies and lacking intelligent data analysis capabilities. Summary of the Invention

[0004] To address the aforementioned issues, the present invention aims to provide a method and system for intelligent monitoring of power distribution networks based on a data platform and edge computing. By integrating edge computing and a data platform, and through multi-level detection and comprehensive analysis, the method can promptly detect and predict anomalies in the network, providing a reliable guarantee for the intelligent management and optimization of power distribution networks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for intelligent monitoring of power distribution networks based on data middleware and edge computing includes the following steps:

[0007] Step S1: Deploy sensors in the power distribution network to collect key data in real time, including current, voltage, and temperature, and perform noise reduction and normalization processing on the collected data.

[0008] Step S2: At the edge, use a lightweight GMM model to perform preliminary anomaly detection on the processed data, and transmit the preprocessed data and preliminary anomaly detection results to the data platform;

[0009] Step S3: The data platform centrally stores the preprocessed data from the edge nodes, and based on the preliminary anomaly detection results, it uses a variational autoencoder for secondary detection to obtain the final detection results;

[0010] Step S4: Construct a causal relationship diagram of the power distribution network, and based on the final detection results, perform anomaly analysis and obtain the anomaly analysis results;

[0011] Step S5: Based on the final detection results and anomaly analysis results, generate a maintenance report and send the maintenance report to the maintenance personnel in a visual format.

[0012] Furthermore, the collected data undergoes denoising and normalization processing as follows:

[0013] For the data x(c) collected by the c-th sensor, noise reduction is performed using a low-pass filter:

[0014] y(c)=αy(c-1)+(1-α)x(c)

[0015] Where α is the smoothing factor of the filter, and y(c) is the output data of the k-th sensor;

[0016] And it is processed using Min-Max normalization:

[0017]

[0018] Where, x(c) min and x(c) max Find the minimum and maximum values ​​of the c-th sensor, respectively.

[0019] Furthermore, step S2 specifically involves:

[0020] Step S21: Based on the denoised and normalized dataset X = {x1, x2, ..., x...} i ,...,x N}, each data point x i It is a three-dimensional vector, representing the normalized current, voltage, and temperature, respectively;

[0021] Step S22: Suppose that data point x is generated by a linear combination of K multivariate Gaussian distributions, and determine the optimal number of Gaussian components K in the GMM model based on BIC. The probability density of data point x is:

[0022]

[0023] Step S23: Iteratively update the GMM model parameters θ using the EM algorithm;

[0024] Step S24: Calculate the log-likelihood probability for each data point;

[0025]

[0026] Where, π k For mixed weights, μ k The mean; ∑ k It is the covariance matrix;

[0027] Step S25: Set a threshold ∈, and mark data points below the threshold to obtain preliminary anomaly detection results:

[0028]

[0029] Where, label i Let i represent data point i.

[0030] Furthermore, the optimal number of Gaussian components K in the GMM model is determined based on BIC, as follows:

[0031] ① Determine the range of candidate K values;

[0032] ② Initialize the variables that store the optimal BIC value and the corresponding K value;

[0033] ③ For each candidate K value, perform the following steps:

[0034] The dataset is fitted using the GMM model to obtain the maximum likelihood estimate ln(L);

[0035] Calculate the number of model parameters k ：

[0036]

[0037] Where d is the dimension of the data (d = 3 in this embodiment);

[0038] Calculate the BIC value:

[0039] BIC = k ln(N) - 2 ln(L)

[0040] If the current BIC value is less than the stored best BIC value, then update the best BIC value and the corresponding K value;

[0041] ④ Select K with the smallest BIC value as the final number of Gaussian components.

[0042] Furthermore, the EM algorithm is used to iteratively update the GMM model parameters θ, as follows:

[0043] Initialize GMM model parameters:

[0044]

[0045] Repeat the following steps until the model converges:

[0046] A. Calculate the posterior probability that each data point belongs to each Gaussian component:

[0047]

[0048] in, It is a Gaussian probability density function;

[0049] B. Update model parameters using responsibility values:

[0050]

[0051] The superscript T indicates transpose.

[0052] Furthermore, for each newly arrived data point x new The parameters of the lightweight GMM model are updated in real time using the Bayesian update formula:

[0053] Update of mixed weights:

[0054]

[0055] Where, γ k It is the responsibility value corresponding to the new data point xnew; γ k Let represent the posterior probability that a new data point belongs to the k-th Gaussian component, calculated using the following formula:

[0056]

[0057] Bayesian update of the mean vector:

[0058]

[0059] Bayesian update of the covariance matrix:

[0060]

[0061] Furthermore, a variational autoencoder includes an encoder and a decoder, as detailed below:

[0062] The encoder maps the initial anomaly detection result x′ to the mean u(x′) and variance σ of the latent variable z. 2 The distribution of (x′), where the latent variables follow a multidimensional normal distribution:

[0063] q φ (z|x′)=Ν(z;u(x′),σ 2 (x′));

[0064] Where N() represents the normal distribution, and φ is the encoder parameter;

[0065] The decoder reconstructs the input data from the latent variable z. Output the generated probability distribution:

[0066]

[0067] Where θ′ are the decoder parameters;

[0068] Optimization is performed using gradient descent, and randomness is removed from the network using reparameterization techniques.

[0069] z=u(x′)+σ(x′)·ε,ε~N(0,I);

[0070] After obtaining the deterministic z, the decoder generates reconstructed data.

[0071] The training objective of the variational autoencoder is to maximize the ELBO, i.e., minimize the following loss function:

[0072] L = L rec +L KL ;

[0073]

[0074] Among them, L rec Reconstruction error is used to measure the difference between the input data x′ and the reconstructed data. Differences between: L KL KL divergence measure is used to measure the output distribution and prior distribution of a quantity encoder.

[0075] Furthermore, the Peter-Clark algorithm, the GIES algorithm, and domain knowledge are used to construct a causal relationship graph for the power distribution network, as detailed below:

[0076] An undirected skeleton graph is constructed based on the conditional independence test, and the undirected edges are oriented.

[0077] Initialize the undirected skeleton graph G:

[0078] For each pair of variables Xi and Xj, if Xi⊥Xj∣S, then delete the edge Xi-Xj;

[0079] S is a set of variables, initially empty, with variables gradually added;

[0080] Where Xi⊥Xj∣S represents the independence test of Xi and Xj under all possible condition sets S;

[0081] Edge orientation:

[0082] For node pairs satisfying Xi-Xj-Xk, if Then Xi→Xj←Xk; where, This indicates that, given Xj, Xi and Xk still have a dependency relationship;

[0083] For the triangle Xi-Xj-Xk→Xj, if Xi→Xj←Xk, then Xi→Xj←Xk.

[0084] Constructing a causal graph through greedy search and data intervention:

[0085] A. Start with the initial undirected graph Ginitial;

[0086] B. In the candidate edge set, optimize the graph's score (e.g., BIC score) by adding / removing / reversing edges.

[0087] C. Use local search methods to progressively optimize the directed acyclic graph structure.

[0088] D. Correct the direction of the edges according to the intervention data to form the final causal graph.

[0089] Furthermore, based on the final detection results, anomaly analysis is performed to obtain the anomaly analysis results, as follows:

[0090] Integrate anomaly detection results into a causal relationship graph to identify the causal paths of anomalies;

[0091] Extract causal paths containing outliers from the causal relationship graph and analyze the impact of these paths on the outcome variables:

[0092]

[0093] Wherein, path p k Includes outliers;

[0094] Analyze the critical path containing outliers and calculate the direct and indirect impacts of the outliers on the outcome variables:

[0095]

[0096] Where β is the path coefficient;

[0097] Quantify the impact of each outlier on the outcome variable and adjust it based on the path coefficient:

[0098]

[0099] Where δ is the anomaly gain factor, which reflects the impact of anomalies on path coefficients;

[0100] The corrected path coefficients are summarized to obtain the critical path and causal effect that include the anomaly impact, i.e., the anomaly analysis results.

[0101] A power distribution network intelligent monitoring system based on a data middleware platform and edge computing includes a data acquisition unit, an edge computing unit, a data middleware platform, and a mobile unit. The data acquisition unit collects key data in real time, including current, voltage, and temperature, using sensors deployed throughout the power distribution network, and performs noise reduction and normalization on the collected data. The edge computing unit uses a lightweight Gaussian Mixture Model (GMM) to perform preliminary anomaly detection on the processed data and transmits the preprocessed data and preliminary anomaly detection results to the data middleware platform. The data middleware platform centrally stores the preprocessed data from edge nodes and, based on the preliminary anomaly detection results, performs secondary detection using a variational autoencoder to obtain the final detection results. It then uses the Peter-Clark algorithm, the GIES algorithm, and domain knowledge to construct a causal relationship graph of the power distribution network and performs anomaly analysis based on the final detection results. Finally, based on the final detection results and anomaly analysis results, a maintenance report is generated and sent to the mobile terminal.

[0102] The present invention has the following beneficial effects:

[0103] 1. This invention integrates edge computing and a data platform, and through multi-level detection and comprehensive analysis, it can promptly detect and predict anomalies in the network, providing a reliable guarantee for the intelligent management and optimization of the power distribution network;

[0104] 2. This invention utilizes a lightweight Gaussian Mixture Model (GMM) at the edge to perform preliminary anomaly detection on the processed data, effectively improving the accuracy and reliability of subsequent secondary detection. By using BIC to select the most suitable number of Gaussian components, the model can ensure good fitting performance while avoiding overfitting and improving the accuracy of anomaly detection. The GMM model parameters are updated in real time through the Bayesian update formula, which can dynamically adapt to new data, thereby improving its responsiveness to changes in data distribution.

[0105] 3. This invention constructs a causal relationship diagram of the power distribution network, conducts in-depth analysis of abnormal causes based on the detection results, helps identify potential root causes and the scope of impact, and provides detailed background and upstream and downstream impacts of abnormal events through multi-level analysis, supporting more accurate fault location and repair decisions. Attached Figure Description

[0106] Figure 1 This is a flowchart of the method of the present invention;

[0107] Figure 2 This is a schematic diagram of the system architecture in one embodiment of the present invention. Detailed Implementation

[0108] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments:

[0109] refer to Figure 1 In this embodiment, a method for intelligent monitoring of power distribution networks based on data middleware and edge computing is provided, including the following steps:

[0110] Step S1: Deploy sensors in the power distribution network to collect key data in real time, including current, voltage, and temperature, and perform noise reduction and normalization processing on the collected data.

[0111] Step S2: At the edge, use a lightweight GMM model to perform preliminary anomaly detection on the processed data, and transmit the preprocessed data and preliminary anomaly detection results to the data platform;

[0112] Step S3: The data platform centrally stores the preprocessed data from the edge nodes, and based on the preliminary anomaly detection results, it uses a variational autoencoder for secondary detection to obtain the final detection results;

[0113] Step S4: Construct a causal relationship diagram of the power distribution network, and based on the final detection results, perform anomaly analysis and obtain the anomaly analysis results;

[0114] Step S5: Based on the final detection results and anomaly analysis results, generate a maintenance report and send the maintenance report to the maintenance personnel in a visual format.

[0115] In this embodiment, the noise reduction and normalization processing of the collected data specifically involves:

[0116] For the data x(c) collected by the c-th sensor, noise reduction is performed using a low-pass filter:

[0117] y(c)=αy(c-1)+(1-α)x(c)

[0118] Where α is the smoothing factor of the filter, and y(c) is the output data of the k-th sensor;

[0119] And it is processed using Min-Max normalization:

[0120]

[0121] Where, x(c) min and x(c) max Find the minimum and maximum values ​​of the c-th sensor, respectively.

[0122] In this embodiment, step S2 specifically includes:

[0123] Step S21: Based on the denoised and normalized dataset X = {x1, x2, ..., x...} i ,...,x N}, each data point x iIt is a three-dimensional vector, representing the normalized current, voltage, and temperature, respectively;

[0124] Step S22: Suppose that data point x is generated by a linear combination of K multivariate Gaussian distributions, and determine the optimal number of Gaussian components K in the GMM model based on BIC. The probability density of data point x is:

[0125]

[0126] Step S23: Iteratively update the GMM model parameters θ using the EM algorithm;

[0127] Step S24: Calculate the log-likelihood probability for each data point;

[0128]

[0129] Where, π k For mixed weights, μ k Σ is the mean; k It is the covariance matrix;

[0130] Step S25: Set a threshold ∈, and mark data points below the threshold to obtain preliminary anomaly detection results:

[0131]

[0132] Where, label i Let i represent data point i.

[0133] In this embodiment, the optimal number of Gaussian components K in the GMM model is determined based on BIC, as follows:

[0134] ① Determine the range of candidate K values;

[0135] ② Initialize the variables that store the optimal BIC value and the corresponding K value;

[0136] ③ For each candidate K value, perform the following steps:

[0137] The dataset is fitted using the GMM model to obtain the maximum likelihood estimate ln(L);

[0138] Calculate the number of model parameters k:

[0139]

[0140] Where d is the dimension of the data (d = 3 in this embodiment);

[0141] Calculate the BIC value:

[0142] BIC = k ln(N) - 2 ln(L)

[0143] If the current BIC value is less than the stored best BIC value, then update the best BIC value and the corresponding K value;

[0144] ④ Select K with the smallest BIC value as the final number of Gaussian components.

[0145] In this embodiment, the EM algorithm is used to iteratively update the GMM model parameters θ, as follows:

[0146] Initialize GMM model parameters:

[0147]

[0148] Repeat the following steps until the model converges:

[0149] A. Calculate the posterior probability that each data point belongs to each Gaussian component:

[0150]

[0151] in, It is a Gaussian probability density function;

[0152] B. Update model parameters using responsibility values:

[0153]

[0154] The superscript T indicates transpose.

[0155] In this embodiment, for each newly arrived data point x new The parameters of the lightweight GMM model are updated in real time using the Bayesian update formula:

[0156] Update of mixed weights:

[0157]

[0158] Where, γ k It is a new data point xnew Corresponding responsibility value; γ k Let represent the posterior probability that a new data point belongs to the k-th Gaussian component, calculated using the following formula:

[0159]

[0160] Bayesian update of the mean vector:

[0161]

[0162] Bayesian update of the covariance matrix:

[0163]

[0164] In this embodiment, the variational autoencoder includes an encoder and a decoder, as detailed below:

[0165] The encoder maps the initial anomaly detection result x′ to the mean u(x′) and variance σ of the latent variable z. 2 The distribution of (x′), where the latent variables follow a multidimensional normal distribution:

[0166] q φ (z|x′)=Ν(z;u(x′),σ 2 (x′));

[0167] Where N() represents the normal distribution, and φ is the encoder parameter;

[0168] The decoder reconstructs the input data from the latent variable z. Output the generated probability distribution:

[0169]

[0170] Where θ′ are the decoder parameters;

[0171] Optimization is performed using gradient descent, and randomness is removed from the network using reparameterization techniques.

[0172] z=u(x′)+σ(x′)·ε,ε~N(0,I);

[0173] After obtaining the deterministic z, the decoder generates reconstructed data.

[0174] The training objective of the variational autoencoder is to maximize the ELBO, i.e., minimize the following loss function:

[0175] L = L rec +L KL ;

[0176]

[0177] Among them, L rec Reconstruction error is used to measure the difference between the input data x′ and the reconstructed data. Differences between: L KL KL divergence measure is used to measure the output distribution and prior distribution of a quantity encoder.

[0178] In this embodiment, the Peter-Clark algorithm, the GIES algorithm, and domain knowledge are used to construct the causal relationship graph of the power distribution network, as detailed below:

[0179] An undirected skeleton graph is constructed based on the conditional independence test, and the undirected edges are oriented.

[0180] Initialize the undirected skeleton graph G:

[0181] For each pair of variables Xi and Xj, if Xi⊥Xj∣S (conditionally independent), then delete the edge Xi-Xj; S is a set of variables, initially empty, and gradually increasing in variables;

[0182] Where Xi⊥Xj∣S represents the independence test of Xi and Xj under all possible condition sets S;

[0183] Edge orientation:

[0184] For node pairs satisfying Xi-Xj-Xk, if Then Xi→Xj←Xk; where, This indicates that, given Xj, Xi and Xk still have a dependency relationship;

[0185] For the triangle Xi-Xj-Xk→Xj, if Xi→Xj←Xk, then Xi→Xj←Xk.

[0186] Constructing a causal graph through greedy search and data intervention:

[0187] A. Start with the initial undirected graph Ginitial;

[0188] B. In the candidate edge set, optimize the graph's score (e.g., BIC score) by adding / removing / reversing edges.

[0189] C. Use local search methods to progressively optimize the directed acyclic graph structure.

[0190] D. Correct the direction of the edges according to the intervention data to form the final causal graph.

[0191] In this embodiment, based on the final detection results, anomaly analysis is performed to obtain the anomaly analysis results, as follows:

[0192] Integrate anomaly detection results into a causal relationship graph to identify the causal paths of anomalies;

[0193] Extract causal paths containing outliers from the causal relationship graph and analyze the impact of these paths on the outcome variables:

[0194]

[0195] Wherein, path p k Includes outliers;

[0196] Analyze the critical path containing outliers and calculate the direct and indirect impacts of the outliers on the outcome variables:

[0197]

[0198] Where β is the path coefficient;

[0199] Quantify the impact of each outlier on the outcome variable and adjust it based on the path coefficient:

[0200]

[0201] Where δ is the anomaly gain factor, which reflects the impact of anomalies on path coefficients;

[0202] The corrected path coefficients are summarized to obtain the critical path and causal effect that include the anomaly impact, i.e., the anomaly analysis results.

[0203] refer to Figure 2 In another embodiment, a power distribution network intelligent monitoring system based on a data middle platform and edge computing is also provided. The system comprises a data acquisition unit, an edge computing unit, a data middle platform, and a mobile unit. The data acquisition unit collects key data in real time, including current, voltage, and temperature, using sensors deployed in the power distribution network, and performs noise reduction and normalization processing on the collected data. The edge computing unit uses a lightweight GMM model to perform preliminary anomaly detection on the processed data and transmits the preprocessed data and preliminary anomaly detection results to the data middle platform. The data middle platform centrally stores the preprocessed data from edge nodes and, based on the preliminary anomaly detection results, performs secondary detection using a variational autoencoder to obtain the final detection results. It then uses the Peter-Clark algorithm, the GIES algorithm, and domain knowledge to construct a causal relationship graph of the power distribution network. Based on the final detection results, it performs anomaly analysis to obtain anomaly analysis results. Finally, based on the final detection results and anomaly analysis results, it generates a maintenance report and sends the report to the mobile terminal.

[0204] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0205] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0206] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0207] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0208] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A distribution network intelligent monitoring method based on data center and edge computing, characterized in that: The following steps are involved: Step S1: Arrange sensors in the power distribution network to collect key data in real time, including current, voltage, and temperature, and perform denoising and normalization on the collected data; Step S2: Use the lightweight GMM model at the edge to perform preliminary anomaly detection on the processed data, and transmit the preprocessed data and preliminary anomaly detection results to the data center; Step S3: The data center centrally stores the pre-processed data from the edge nodes, and based on the preliminary anomaly detection results, uses the variational autoencoder for secondary detection to obtain the final detection results; Step S4: construct a causal relationship diagram of the power distribution network, and perform anomaly analysis based on the final detection results to obtain anomaly analysis results; Step S5: Generate a maintenance report based on the final detection results and abnormality analysis results, and send the maintenance report to the maintenance personnel in a visual display.

2. According to claim 1, a distribution network intelligent monitoring method based on data middle station and edge computing is characterized in that: The denoising and normalization processing of the collected data is specifically as follows: For the data x(c) collected by the cth sensor, denoising is performed through a low-pass filter: y(c)=αy(c-1)+(1-α)x(c) Among them, α is the smoothing factor of the filter, and y(c) is the output data of the kth sensor; And use Min-Max normalization: Among them, x(c) min and x(c) max are the minimum and maximum values ​​of the cth sensor, respectively.

3. According to claim 1, a distribution network intelligent monitoring method based on data middle platform and edge computing is characterized in that: The step S2 is specifically as follows: Step S21: Based on the denoised and normalized data set X = {x1, x2, ..., x i ,...,x N }, each data point x i is a three-dimensional vector, representing the normalized current, voltage, and temperature respectively; Step S22: Assume that the data point x is generated by a linear combination of K multivariate Gaussian distributions, and determine the optimal number of Gaussian components K of the GMM model based on BIC. The probability density of the data point x is: Step S23: Use the EM algorithm to iteratively update the GMM model parameters θ; Step S24: Calculate the log-likelihood probability for each data point; Among them, π k is the mixing weight, μ k is the mean; k is the covariance matrix; Step S25: Set a threshold ∈, mark the data points below the threshold, and obtain preliminary anomaly detection results: Among them, label i represents data point i.

4. According to claim 3, a distribution network intelligent monitoring method based on data middle platform and edge computing is characterized in that: The optimal number of Gaussian components K of the GMM model is determined based on BIC as follows: ①Determine the candidate K value range; ② Initialize the variables that store the best BIC value and the corresponding K value; ③For each candidate K value, perform the following steps: Use the GMM model to fit the data set and obtain the maximum likelihood estimate ln(L); Calculate the number of model parameters k ： Wherein, d is the dimension of the data (in this embodiment, d=3); Calculate the BIC value: BIC = k ln(N)-2 ln(L) If the current BIC value is less than the stored optimal BIC value, update the optimal BIC value and the corresponding K value; ④ Select K with the smallest BIC value as the final number of Gaussian components.

5. The method for intelligent monitoring of power distribution network based on data middle platform and edge computing according to claim 4 is characterized in that: The EM algorithm is used to iteratively update the GMM model parameters θ, specifically as follows: Initialize the GMM model parameters: Repeat the following steps until the model converges: A. Calculate the posterior probability that each data point belongs to each Gaussian component: in, is the Gaussian distribution probability density function; B. Update model parameters using responsibility values: The superscript T indicates transpose.

6. A distribution network intelligent monitoring method based on data middle platform and edge computing according to claim 5, characterized in that: For each new data point x new , use the Bayesian update formula to update the lightweight GMM model parameters in real time: Update of mixing weights: Among them, γ k Is a new data point xnew Corresponding responsibility value responsibility value; γ k It represents the posterior probability that the new data point belongs to the kth Gaussian component, and the calculation formula is as follows: Bayesian update of the mean vector: Bayesian update of the covariance matrix:

7. The method for intelligent monitoring of power distribution network based on data middle platform and edge computing according to claim 1 is characterized in that: The variational autoencoder includes an encoder and a decoder, which are as follows: The encoder maps the preliminary anomaly detection result x′ to the mean u(x′) and variance σ of the latent variable z 2 The distribution of (x′), where the latent variable follows a multidimensional normal distribution: q φ (z|x′)=Ν(z;u(x′),σ 2 (x′)); Wherein, Ν() represents normal distribution, φ is the parameter of the encoder; The decoder reconstructs the input data from the latent variables z The output generates a probability distribution: Among them, θ′ is the decoder parameter; Optimize via gradient descent, using the reparameterization trick to remove randomness from the network: z=u(x′)+σ(x′)·ε,ε~N(0,I); After obtaining the deterministic z, the decoder generates the reconstructed data The training goal of the variational autoencoder is to maximize ELBO, that is, to minimize the following loss function: L=L rec +L KL ; Among them, L rec is the reconstruction error, which is used to measure the difference between the input data x′ and the reconstructed data The difference between: L KL is the KL divergence measure, which is used to measure the output distribution of the encoder and the prior distribution.

8. According to claim 1, a distribution network intelligent monitoring method based on data middle platform and edge computing is characterized in that: The Peter-Clark algorithm, GIES algorithm and domain knowledge are used to construct a causal relationship diagram of the distribution network, as follows: Construct an undirected skeleton graph based on conditional independence test and orient the undirected edges; Initialize the undirected skeleton graph G: For each pair of variables Xi and Xj, if Xi⊥Xj|S, delete the edge Xi-Xj; S is a set of variables, which is initially empty and gradually increases in variables; Among them, Xi⊥Xj|S means that Xi and Xj are tested for independence under all possible condition sets S; Orientation of edges: For node pairs that satisfy Xi-Xj-Xk, if Then Xi→Xj←Xk; among them, It means that given Xj, Xi and Xk still have a dependency relationship; The triangle Xi-Xj-Xk→Xj is formed. If Xi→Xj←Xk, then Xi→Xj←Xk; Constructing a causal graph through greedy search and intervening data: A. Start from the initial undirected graph Ginitial; B. In the candidate edge set, optimize the graph score (such as BIC score) by adding / deleting / reversing edges. C. Use local search method to gradually optimize the directed acyclic graph structure. D. Correct the direction of the edge according to the intervening data to form the final causal graph.

9. The method for intelligent monitoring of power distribution network based on data middle platform and edge computing according to claim 8 is characterized in that: Based on the final detection result, an abnormal analysis is performed to obtain an abnormal analysis result, as follows: Integrate the anomaly detection results into the causal relationship diagram to mark the causal path of the anomaly points; Extract causal paths containing outliers from the causal relationship graph and analyze the impact of these paths on the outcome variables: Among them, the path p k Contains outliers; Analyze the critical paths containing outliers and calculate the direct and indirect impact of outliers on the outcome variables: Among them, β is the path coefficient; Quantify the impact of each outlier on the outcome variable and correct it based on the path coefficient: Among them, δ is the abnormal gain factor to reflect the impact of abnormality on the path coefficient; The corrected path coefficients are summarized to obtain the key paths and causal effects including abnormal influences, which are the abnormal analysis results.

10. An intelligent monitoring system for power distribution network based on data center and edge computing, characterized in that: It includes a data acquisition unit, an edge computing unit, a data middle platform and a mobile unit; the data acquisition unit collects key data in real time, including current, voltage, and temperature, through sensors arranged in the power distribution network, and performs denoising and normalization processing on the collected data; The edge computing unit uses a lightweight GMM model to perform preliminary anomaly detection on the processed data, and transmits the preprocessed data and the preliminary anomaly detection results to the data center; the data center centrally stores the preprocessed data from the edge nodes, and based on the preliminary anomaly detection results, uses a variational autoencoder to perform secondary detection to obtain the final detection results, and uses the Peter-Clark algorithm, the GIES algorithm and domain knowledge to construct a causal relationship diagram of the distribution network, and based on the final detection results, performs anomaly analysis to obtain anomaly analysis results; Based on the final detection results and abnormal analysis results, a maintenance report is generated and sent to the mobile terminal.

Images