A centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT

Abstract

This invention discloses a centralized-distributed collaborative fault detection method for distribution networks based on LOF-VAE-GPT. It collects physical and network data from key edge nodes of the distribution network; constructs a physical anomaly detection model for key edge nodes based on the local outlier factor algorithm to detect anomalous physical samples; proposes a network anomaly detection model for key edge nodes based on a variational autoencoder to detect anomalies in network samples of key edge nodes; and trains the GPT model using the detected physical and network anomaly samples to construct a distribution network fault detection model. When the detected outlier value of a node exceeds a threshold, it is determined that a fault has occurred at the edge node of the distribution network. This invention can improve the effectiveness of anomaly identification in physical data, the adaptability to the complex environment of key edge nodes in the distribution network, and the accuracy of fault detection for key nodes in the distribution network, thereby improving the safe operation level of the distribution network.

Description

Technical Field

[0001] This invention relates to the field of power system distribution network fault detection technology, and in particular to a centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT. Background Technology

[0002] Detecting anomalies and faults in critical edge nodes of the distribution network is of great significance. These nodes play a crucial role in power transmission and distribution; any anomalies or faults in these nodes can lead to power outages, affecting users' normal electricity consumption. Existing technologies can reduce outage time and improve power supply reliability by promptly detecting and addressing these issues. Anomalies in edge nodes may be precursors to larger accidents; real-time monitoring and early warning systems allow for timely intervention to prevent escalation and ensure personnel safety and equipment integrity. Automated monitoring systems can quickly locate fault points, reducing on-site troubleshooting time and improving fault handling efficiency. Regular inspection and maintenance can reduce costs associated with emergency fault repairs, extend equipment lifespan, and lower long-term operating costs. Deploying intelligent monitoring and diagnostic systems can promote the automation and intelligence of the distribution network, improving the scientific and advanced nature of distribution network management. Monitoring data can support power system decision-making, such as equipment upgrades, capacity expansion, and dispatching. Edge node monitoring enables better energy management and optimization, improving the overall grid operating efficiency. Stable power supply increases user satisfaction and reduces user complaints caused by power problems.

[0003] However, existing methods for collecting network physical data and detecting anomalies at critical edge nodes in distribution networks still have the following shortcomings: 1) Insufficient real-time performance: Existing methods may not be able to collect and process large amounts of network physical data in real time, resulting in slow response in emergency situations; 2) Accuracy issues: Traditional methods may rely on simple threshold judgments, which are easily affected by noisy data and anomalies, leading to false alarms or missed alarms; 3) Weak generalization ability: Some methods perform poorly when dealing with complex network structures and diverse data types, making it difficult to adapt to different network environments and conditions.

[0004] To address these issues, we designed a centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies in the acquisition of network physical data and anomaly detection of key edge nodes in distribution networks, such as insufficient real-time performance, insufficient accuracy, and weak generalization ability. This invention proposes a centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT, which improves the effectiveness of anomaly identification in physical data, enhances adaptability to the complex environment of key edge nodes in distribution networks, and improves the accuracy of fault detection at key nodes in distribution networks.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A centralized-distributed collaborative fault detection method for distribution networks based on LOF-VAE-GPT includes the following steps:

[0008] Step 1: Collect physical and network data of key edge nodes in the distribution network. Physical data includes current data, voltage data, power data, frequency data, and voltage phase angle data. Network data includes traffic information, log information, and message information.

[0009] Step 2: Based on the local outlier factor algorithm, construct a physical anomaly detection model for key edge nodes of the distribution network, detect abnormal physical samples of key edge nodes of the distribution network, and identify the physical abnormal behavior of key nodes in the distribution network.

[0010] Step 3: Propose a network anomaly detection model for key edge nodes of the distribution network based on variational autoencoder, and perform anomaly detection on network samples of key edge nodes of the distribution network.

[0011] Step 4: Based on the GPT model and the detected physical and network anomaly samples, train the model, calculate the outlier values, and build a distribution network fault detection model. Input the real-time collected physical and network data of key edge nodes of the distribution network into the distribution network fault detection model. When the outlier value of a detected node is greater than the threshold, it is determined that the edge node of the distribution network has failed.

[0012] To further explain, in step 1, the physical and network data of the key edge nodes of the distribution network are collected using the following formula:

[0013] X Gi (t)=[I i (t),U i (t),P i (t),f i (t),θ i (t)] (1)

[0014] X Wi (t)=[L i (t),R i (t),B i(t)] (2)

[0015] In the above formula, X Gi (t) represents the physical data of the critical edge node i of the distribution network collected at time t. i (t) represents the current data of the key edge node i in the distribution network at time t. i (t) represents the voltage data of the critical edge node i in the distribution network at time t. i (t) represents the power data of the key edge node i in the distribution network at time t, f i (t) represents the frequency data of the key edge node i in the distribution network at time t, θ i (t) represents the voltage phase angle data of the key edge node i in the distribution network at time t; X Wi (t) represents the network data collected at time t for the critical edge node i of the distribution network. i (t) represents the traffic information of the critical edge node i of the distribution network collected at time t, R i (t) represents the log information of the critical edge node i of the distribution network collected at time t, B i (t) represents the message information of the critical edge node i of the distribution network collected at time t.

[0016] To further explain, in step 2, a physical anomaly detection model for key edge nodes of the distribution network is constructed based on the local outlier factor algorithm. This model detects abnormal physical samples of key edge nodes in the distribution network and identifies abnormal physical behaviors of key nodes in the distribution network. The steps include:

[0017] Step 21: For each physical sample of a critical edge node in the distribution network, calculate the k-distance between physical samples using the k-distance calculation formula:

[0018] d k (i,j)=(I i (t)-I j (t)) 2 +(U i (t)-U j (t)) 2 +(P i (t)-P j (t)) 2

[0019] +(f i (t)-f j (t)) 2 +(θ i (t)-θ j (t)) 2 (3)

[0020] In the formula, d k(i,j) represents the physical sample distance between critical edge node i and node j in the distribution network, and k represents the preset parameter; j (t) represents the current data of the key edge node j of the distribution network at time t, U j (t) represents the voltage data of the key edge node j of the distribution network at time t. j (t) represents the power data of the key edge node j of the distribution network at time t, f j (t) represents the frequency data of the key edge node j of the distribution network at time t, θ j (t) represents the voltage phase angle data of the key edge node j of the distribution network at time t.

[0021] Step 22: For the physical samples of critical edge nodes in the distribution network, calculate the local reachability density ρ of the physical samples using the formula for calculating the local reachability density of critical edge nodes in the distribution network. k (i):

[0022]

[0023] In the formula, N k (i) is the set of the k nearest neighbors of the physical sample;

[0024] Step 23: Calculate the Local Outlier Factor (LOF) for each physical sample. k (i):

[0025]

[0026] In the formula, ρ k (j) represents the local reachability density of the physical samples of the critical edge node j in the distribution network;

[0027] Step 24: Perform anomaly detection on the physical samples of key edge nodes of the distribution network, and determine whether the physical samples of key edge nodes of the distribution network are abnormal based on the local outlier factor:

[0028] LOF k (i)≥δ k (6)

[0029] In the formula, δ k The physical sample anomaly threshold, when LOF k (i)≥δ k At that time, the physical samples of the key edge nodes of the distribution network are marked as abnormal, and the abnormal physical behavior of the key nodes in the distribution network is identified.

[0030] To further explain, in step 3, a network anomaly detection model based on variational autoencoders for key edge nodes of the distribution network is proposed. Anomaly detection is performed on network samples of key edge nodes of the distribution network using the following steps:

[0031] Step 31: Encode the network samples based on the network sample encoding function of the key edge node of the distribution network using the variational autoencoder. The expression is as follows:

[0032] Z w,i =f VAE,BM (X W,i (7)

[0033] In the formula, Z w,i f represents the potential sample after encoding of the critical edge node i in the distribution network. VAE,BM () represents the network sample encoding function for the critical edge node i of the distribution network based on a variational autoencoder, X W,i This is a network sample for a key edge node i in the distribution network.

[0034] Step 32, the approximate posterior distribution of the key edge nodes of the distribution network based on the variational autoencoder is expressed by the following formula:

[0035]

[0036] μ W =f1(X W,i (9)

[0037]

[0038] In the formula, q(Z) w,i |X W,i Let N be the approximate posterior distribution value of the critical edge node of the distribution network based on the variational autoencoder. VAE () represents the approximate posterior distribution function of the critical edge node network of the distribution network based on variational autoencoder, μ W This represents the mean of network samples based on a neural network model. Let f1() and f2() be the variance based on the neural network model, and f1() and f2() be the model functions of the BP neural network and the convolutional neural network, respectively.

[0039] Step 33: Based on the loss function during the training process of the variational autoencoder model, the total loss of the variational autoencoder is minimized using gradient descent to train the variational autoencoder model. After training, the network samples to be detected are input into the trained variational autoencoder model, which outputs latent samples. The latent samples are then decoded to obtain the reconstructed network samples of the key edge nodes of the distribution network.

[0040] L C =f E [log(P(X W,i |Z W,i )]+f D [q(Z w,i |X W,i )||P(ZW,i (11)

[0041] X' W,i =f VAE,BM (Z W,i (12)

[0042] In the formula, L C P(X) is the loss function during the training process of the variational autoencoder model. W,i |Z W,i ) represents the network sample distribution, based on X W,i The sample distribution P(X) W,i ) and Z W,i The sample distribution P(Z) W,i The ratio of f is used to calculate f. E []、f D [ ] represent the reconstruction loss function and the KL divergence loss function, respectively, X' W,i This is a network sample of key edge nodes in the reconstructed distribution network;

[0043] Step 34: Construct a network anomaly detection model for key edge nodes of the distribution network, calculate the differences between the original data and the reconstructed network samples of key edge nodes of the distribution network, and perform anomaly detection on the network samples of key edge nodes of the distribution network.

[0044] S W,i =f VAE,C (X W,i ,X' W,i (13)

[0045] S W,i >δ W (14)

[0046] In the formula, S W,i For anomaly scoring of network samples, f VAE,C () represents the scoring calculation model function, δ W The threshold for network score anomalies is S. W,i >δ W At that time, the network samples of the critical edge nodes of the distribution network were determined to be abnormal.

[0047] To further explain, in step 4, a distribution network fault detection model is constructed by training the GPT model and the detected physical and network anomaly samples. The expression is as follows:

[0048]

[0049] Y i ≥δ GPT (16)

[0050] In the formula, Y iFor distribution network fault detection scoring, f GPT () represents the function for calculating the distribution network fault detection score. For the detected physically abnormal samples, For the detected network anomaly samples, δ GPT The distribution network fault detection scoring threshold is set; real-time collected physical and network data of key edge nodes of the distribution network are input into the distribution network fault detection model, when Y i ≥δ GPT If this occurs, it indicates that a fault has occurred at the edge node of the distribution network.

[0051] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention, based on the Local Outlier Factor (LOF) algorithm and the Variational Autoencoder (VAE) algorithm, performs anomaly detection on the physical and network information data of key edge nodes in the distribution network. Based on physical and network anomaly sample data, the GPT model is trained to construct a distribution network fault detection model, detecting whether faults exist at key edge nodes. This effectively overcomes the shortcomings of existing technologies in the acquisition of network physical data and anomaly detection of key edge nodes in the distribution network, such as insufficient real-time performance, insufficient accuracy, and weak generalization ability. Using the collaborative distribution network fault detection method proposed in this invention can improve the effectiveness of anomaly identification in physical data, enhance adaptability to the complex environment of key edge nodes in the distribution network, improve the accuracy of fault detection at key nodes in the distribution network, and improve the safe operation level of the distribution network. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT proposed in this invention. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0054] The centralized-distributed collaborative distribution network fault detection method proposed in this embodiment first collects network and physical data of key edge nodes of the distribution network, then uses a physical anomaly detection model of key edge nodes of the distribution network based on the Local Outlier Factor (LOF) algorithm, a network anomaly detection model of key edge nodes of the distribution network based on the Variational Autoencoder (VAE) algorithm, and a distribution network fault detection model based on the GPT model to detect outliers of nodes and determine that the edge nodes of the distribution network have failed.

[0055] Specifically, refer to Figure 1The following section details a centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT, which mainly includes the following steps:

[0056] Step 1: Collect physical and network data from key edge nodes of the distribution network. Physical data includes current data, voltage data, power data, frequency data, and voltage phase angle data. Network data includes traffic information, log information, and message information.

[0057] The physical and network data of key edge nodes in the distribution network are collected and expressed using the following formula:

[0058] X Gi (t)=[I i (t),U i (t),P i (t),f i (t),θ i (t)] (1)

[0059] X Wi (t)=[L i (t),R i (t),B i (t)] (2)

[0060] In the above formula, X Gi (t) represents the physical data of the critical edge node i of the distribution network collected at time t. i (t) represents the current data of the key edge node i in the distribution network at time t. i (t) represents the voltage data of the critical edge node i in the distribution network at time t. i (t) represents the power data of the key edge node i in the distribution network at time t, f i (t) represents the frequency data of the key edge node i in the distribution network at time t, θ i (t) represents the voltage phase angle data of the key edge node i in the distribution network at time t; X Wi (t) represents the network data collected at time t for the critical edge node i of the distribution network. i (t) represents the traffic information of the critical edge node i of the distribution network collected at time t, R i (t) represents the log information of the critical edge node i of the distribution network collected at time t, B i (t) represents the message information of the critical edge node i of the distribution network collected at time t.

[0061] Step 2: Based on the local outlier factor algorithm, construct a physical anomaly detection model for critical edge nodes of the distribution network, detect abnormal physical samples of critical edge nodes, and identify abnormal physical behaviors of critical nodes in the distribution network. Specifically, this includes the following steps:

[0062] Step 21: For each physical sample of a critical edge node in the distribution network, calculate the k-distance between physical samples using the k-distance calculation formula:

[0063] d k (i,j)=(I i (t)-I j (t)) 2 +(U i (t)-U j (t)) 2 +(P i (t)-P j (t)) 2

[0064] +(f i (t)-f j (t)) 2 +(θ i (t)-θ j (t)) 2 (3)

[0065] In the formula, d k (i,j) represents the physical sample distance between critical edge node i and node j in the distribution network, and k represents the preset parameter; j (t) represents the current data of the key edge node j of the distribution network at time t, U j (t) represents the voltage data of the key edge node j of the distribution network at time t. j (t) represents the power data of the key edge node j of the distribution network at time t, f j (t) represents the frequency data of the key edge node j of the distribution network at time t, θ j (t) represents the voltage phase angle data of the key edge node j of the distribution network at time t.

[0066] Step 22: For the physical samples of critical edge nodes in the distribution network, calculate the local reachability density ρ of the physical samples using the formula for calculating the local reachability density of critical edge nodes in the distribution network. k (i):

[0067]

[0068] In the formula, N k (i) is the set of the k nearest neighbors of the physical sample;

[0069] Step 23: Calculate the Local Outlier Factor (LOF) for each physical sample. k (i):

[0070]

[0071] In the formula, ρ k (j) represents the local reachability density of the physical samples of the critical edge node j of the distribution network.

[0072] Step 24: Perform anomaly detection on the physical samples of key edge nodes of the distribution network, and determine whether the physical samples of key edge nodes of the distribution network are abnormal based on the local outlier factor:

[0073] LOF k (i)≥δ k (6)

[0074] In the formula, δ k The physical sample anomaly threshold, when LOF k (i)≥δ k At that time, the physical samples of the key edge nodes of the distribution network are marked as abnormal, and the abnormal physical behavior of the key nodes in the distribution network is identified.

[0075] Step 3 proposes a network anomaly detection model for key edge nodes of the distribution network based on variational autoencoders to detect anomalies in network samples of key edge nodes of the distribution network.

[0076] The following steps are used:

[0077] Step 31: Encode the network samples based on the network sample encoding function of the key edge node of the distribution network using the variational autoencoder. The expression is as follows:

[0078] Z w,i =f VAE,BM (X W,i (7)

[0079] In the formula, Z w,i f represents the potential sample after encoding of the critical edge node i in the distribution network. VAE,BM () represents the network sample encoding function for the critical edge node i of the distribution network based on a variational autoencoder, X W,i This is a network sample for a key edge node i in the distribution network.

[0080] Step 32, the approximate posterior distribution of the key edge nodes of the distribution network based on the variational autoencoder is expressed by the following formula:

[0081]

[0082] μ W =f1(X W,i (9)

[0083]

[0084] In the formula, q(Z) w,i |X W,iLet N be the approximate posterior distribution value of the critical edge node of the distribution network based on the variational autoencoder. VAE () represents the approximate posterior distribution function of the critical edge node network of the distribution network based on variational autoencoder, μ W This represents the mean of network samples based on a neural network model. Let f1() and f2() be the variance based on the neural network model, and f1() and f2() be the model functions of the BP neural network and the convolutional neural network, respectively.

[0085] Step 33: Based on the loss function during the training process of the variational autoencoder model, the total loss of the variational autoencoder (the sum of reconstruction loss and KL divergence loss) is minimized using gradient descent. The variational autoencoder model is then trained. After training, the network samples to be detected are input into the trained variational autoencoder model, which outputs latent samples. The latent samples are then decoded to obtain the reconstructed network samples of the key edge nodes of the distribution network.

[0086] L C =f E [log(P(X W,i |Z W,i )]+f D [q(Z w,i |X W,i )||P(Z W,i (11)

[0087] X' W,i =f VAE,BM (Z W,i (12)

[0088] In the formula, L C P(X) is the loss function during the training process of the variational autoencoder model. W,i |Z W,i ) represents the network sample distribution, based on X W,i The sample distribution P(X) W,i ) and Z W,i The sample distribution P(Z) W,i The ratio of f is used to calculate f. E []、f D [ ] represent the reconstruction loss function and the KL divergence loss function, respectively, X' W,i This is a network sample of key edge nodes in the reconstructed distribution network.

[0089] Step 34: Construct a network anomaly detection model for key edge nodes of the distribution network, calculate the differences between the original data and the reconstructed network samples of key edge nodes of the distribution network, and perform anomaly detection on the network samples of key edge nodes of the distribution network.

[0090] S W,i =f VAE,C (XW,i ,X' W,i (13)

[0091] S W,i >δ W (14)

[0092] In the formula, S W,i For anomaly scoring of network samples, f VAE,C () represents the scoring calculation model function, δ W The threshold for network score anomalies is S. W,i >δ W At that time, the network samples of the critical edge nodes of the distribution network were determined to be abnormal.

[0093] Step 4: Based on the GPT model, analyze the physical anomaly samples detected in Step 2. and the network anomaly samples detected in step 3 The model is trained based on the GPT model and detected physical and network anomaly samples to construct a distribution network fault detection model, the expression of which is as follows:

[0094]

[0095] Y i ≥δ GPT (16)

[0096] In the formula, Y i For distribution network fault detection scoring, f GPT () represents the function for calculating the distribution network fault detection score. For the detected physically abnormal samples, For the detected network anomaly samples, δ GPT The threshold for distribution network fault detection scoring.

[0097] The real-time collected physical and network data of key edge nodes in the distribution network are input into the distribution network fault detection model. When Y i ≥δ GPT If this occurs, it indicates that a fault has occurred at the edge node of the distribution network.

[0098] By following the above steps, effective detection of faults in critical edge nodes of the distribution network can be achieved.

[0099] This invention utilizes the Local Outlier Factor (LOF) algorithm and the Variational Autoencoder (VAE) algorithm to detect anomalies in the physical and network information of critical edge nodes in distribution networks. It employs the GPT model to detect both physical and network anomalies, thereby identifying faults in distribution network nodes. This improves the effectiveness of anomaly identification in physical data, enhances adaptability to the complex environments of critical edge nodes in distribution networks, increases the accuracy of fault detection at critical nodes, and ultimately improves the overall safety and operational level of the distribution network.

[0100] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT, characterized in that, Includes the following steps: Step 1: Collect physical and network data of key edge nodes in the distribution network. Physical data includes current data, voltage data, power data, frequency data, and voltage phase angle data. Network data includes traffic information, log information, and message information. Step 2: Based on the local outlier factor algorithm, construct a physical anomaly detection model for key edge nodes of the distribution network, detect abnormal physical samples of key edge nodes of the distribution network, calculate the local outlier factor of each physical sample, and determine whether the physical samples of key edge nodes of the distribution network are abnormal based on the local outlier factor, thereby identifying the physical abnormal behavior of key nodes in the distribution network. Step 3: Encode the network samples of key edge nodes of the distribution network based on the variational autoencoder network sample encoding function. Based on the loss function during the training process of the variational autoencoder model, minimize the total loss of the variational autoencoder using gradient descent method, train the variational autoencoder model, output latent samples, and decode the latent samples to obtain the reconstructed network samples of key edge nodes of the distribution network. Construct a network anomaly detection model for key edge nodes of the distribution network, calculate the difference between the original data and the reconstructed network samples of key edge nodes of the distribution network, and perform anomaly detection on the network samples of key edge nodes of the distribution network. Step 4: Based on the GPT model and the detected physical and network anomaly samples, train and build a distribution network fault detection model. Input the real-time collected physical and network data of key edge nodes of the distribution network into the distribution network fault detection model. When the detected abnormal value of a node is greater than the threshold, it is determined that the edge node of the distribution network has failed. The distribution network fault detection model is constructed as follows: （15） （16） In the formula, For distribution network fault detection scoring, This represents the function for calculating the distribution network fault detection score. For the detected physically abnormal samples, For detected network anomaly samples, The distribution network fault detection scoring threshold is set; real-time collected physical and network data of key edge nodes of the distribution network are input into the distribution network fault detection model. If this occurs, it indicates that a fault has occurred at the edge node of the distribution network.

2. The centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT according to claim 1, characterized in that, In step 1, the physical and network data of the key edge nodes of the distribution network are collected using the following formula: （1） （2） In the above formula, The physical data of the key edge node i of the distribution network at time t are collected. The data collected are the current data of the key edge node i in the distribution network at time t. The data collected are the voltage data of the key edge node i in the distribution network at time t. The power data of the key edge node i in the distribution network at time t is collected. The frequency data of the key edge node i in the distribution network at time t are collected. The voltage phase angle data of the key edge node i of the distribution network at time t is collected. For the network data of the key edge node i of the distribution network collected at time t, The traffic information of the key edge node i of the distribution network at time t is collected. The log information of the key edge node i of the distribution network at time t is collected. The message information of the critical edge node i of the distribution network at time t is collected.

3. The centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT according to claim 2, characterized in that, In step 2, a physical anomaly detection model for key edge nodes of the distribution network is constructed based on the local outlier factor algorithm. This model detects abnormal physical samples of key edge nodes and identifies abnormal physical behaviors of key nodes in the distribution network. The steps include: Step 21: For each physical sample of a critical edge node in the distribution network, calculate the k-distance between physical samples using the k-distance calculation formula: （3） In the formula, This represents the physical sample distance between key edge node i and node j in the distribution network, where k represents a preset parameter. The data collected are the current data of the key edge node j of the distribution network at time t. The voltage data of the key edge node j of the distribution network at time t is collected. The power data of the key edge node j of the distribution network at time t is collected. For the frequency data of key edge node j of the distribution network at time t, The voltage phase angle data of the key edge node j of the distribution network at time t is collected; Step 22: For the physical samples of critical edge nodes in the distribution network, calculate the local reachability density of the physical samples using the formula for calculating the local reachability density of critical edge nodes in the distribution network. : （4） In the formula, It is the set of the k nearest neighbors of the physical sample; Step 23: Calculate the local outlier factor for each physical sample. : （5） In the formula, Let J be the local reachability density of the physical samples of the critical edge node j in the distribution network. Step 24: Perform anomaly detection on the physical samples of key edge nodes of the distribution network, and determine whether the physical samples of key edge nodes of the distribution network are abnormal based on the local outlier factor: （6） In the formula, The physical sample anomaly threshold, when At that time, the physical samples of the key edge nodes of the distribution network are marked as abnormal, and the abnormal physical behavior of the key nodes in the distribution network is identified.

4. The centralized-distributed collaborative distribution network fault detection method based on LOF-VAE-GPT according to claim 1, characterized in that, In step 3, a network anomaly detection model based on variational autoencoders for key edge nodes of the distribution network is proposed. Anomaly detection is performed on network samples of key edge nodes of the distribution network using the following steps: Step 31: Encode the network samples based on the network sample encoding function of the key edge node of the distribution network using the variational autoencoder. The expression is as follows: （7） In the formula, This represents the potential sample encoded by the critical edge node i in the distribution network. Here is the network sample encoding function for key edge node i in the distribution network based on variational autoencoder. This is a network sample for a key edge node i in the distribution network. Step 32, the approximate posterior distribution of the key edge nodes of the distribution network based on the variational autoencoder is expressed by the following formula: （8） （9） （10） In the formula, This represents the approximate posterior distribution value of the critical edge node of the distribution network based on a variational autoencoder. This is the approximate posterior distribution function of the critical edge node network of the distribution network based on variational autoencoder. This represents the mean of network samples based on a neural network model. For variance based on neural network models, , These are the model functions for a backpropagation neural network and a convolutional neural network, respectively. Step 33: Based on the loss function during the training process of the variational autoencoder model, the total loss of the variational autoencoder is minimized using gradient descent to train the variational autoencoder model. After training, the network samples to be detected are input into the trained variational autoencoder model, which outputs latent samples. The latent samples are then decoded to obtain the reconstructed network samples of the key edge nodes of the distribution network. （11） （12） In the formula, This is the loss function during the training process of the variational autoencoder model; For network sample distribution, , These are the reconstruction loss function and the KL divergence loss function, respectively. This is a network sample of key edge nodes in the reconstructed distribution network; Step 34: Construct a network anomaly detection model for key edge nodes of the distribution network, calculate the differences between the original data and the reconstructed network samples of key edge nodes of the distribution network, and perform anomaly detection on the network samples of key edge nodes of the distribution network. （13） （14） In the formula, For anomaly scoring of network samples, For the scoring calculation model function, The threshold for network score anomalies, when At that time, the network samples of the critical edge nodes of the distribution network were determined to be abnormal.

Images