Microgrid fault detection method and system based on traveling wave distance measurement and distributed edge calculation

Abstract

The invention discloses a micro-grid fault detection method and system based on traveling wave distance measurement and distributed edge calculation. The method comprises the following steps: searching and detecting traveling wave arrival time based on a neural architecture; the historical data and currently detected power grid data are analyzed and modeled through a GPT model, fault feature vectors are extracted, and the currently detected power grid data comprise the propagation speed of traveling waves and the detected traveling wave arrival time; the edge node independently calculates a local fault position based on a fault feature vector extracted by a GPT model; and carrying out weighted integration on the local fault positions calculated by all the edge nodes to obtain a final fault position. According to the invention, the accuracy of traveling wave arrival time prediction and the accuracy of micro-grid fault location are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power equipment, and in particular to a microgrid fault detection method and system based on traveling wave ranging and distributed edge computing. Background Art

[0002] Effective detection of microgrid faults can timely handle faults and improve the safe operation level of microgrids. However, the existing methods still have the following shortcomings: 1) Limited detection accuracy. Existing methods often find it difficult to accurately detect the fault location in a microgrid environment, especially when facing complex network topology structures. The accuracy is not high enough. Due to the small scale and strong dynamics of microgrids, the traditional methods have large errors. 2) Poor real-time performance. Traditional fault detection methods require a lot of data processing and complex calculations. Especially in a distributed environment, it takes a long time to aggregate data to the center for analysis, which cannot meet the real-time requirements. 3) Insufficient anti-interference ability. Microgrids usually have more distributed energy access, frequent power fluctuations, and are easily interfered by the external environment (such as noise and load fluctuations), resulting in reduced reliability of fault detection methods. 4) Unable to fully utilize edge computing: Traditional fault detection methods mostly rely on centralized data processing and cannot fully utilize the potential of distributed edge computing. Edge computing can be analyzed at a location close to the data source, but the existing methods are not efficient in utilizing such computing resources. 5) Lack of in-depth analysis of traveling wave characteristics: Existing traveling wave fault detection methods mostly use fixed thresholds, which are difficult to dynamically adjust according to actual conditions and cannot effectively capture traveling wave changes in microgrids. Summary of the invention

[0003] In order to solve the above problems, the present invention proposes a microgrid fault detection method and system based on traveling wave ranging and distributed edge computing, which can improve the accuracy of microgrid fault location.

[0004] In order to achieve the above object, the present invention is implemented by the following technical solutions:

[0005] The microgrid fault detection method based on traveling wave ranging and distributed edge computing of the present invention includes the following operations:

[0006] Detecting the arrival time of traveling waves based on neural architecture search, including: using sensors to measure the voltage signals of each node and extracting time series features; using neural architecture search algorithms to automatically find the best neural network architecture based on the time series features; predicting the arrival time of traveling waves based on the best neural network architecture, and dynamically adjusting the threshold of the arrival time of traveling waves according to the predicted arrival time of traveling waves; using the threshold of the arrival time of traveling waves to finally detect the arrival time of traveling waves;

[0007] The GPT model is used to analyze and model the historical data and the currently detected power grid data to extract the fault feature vector, wherein the currently detected power grid data includes the propagation speed of the traveling wave and the detected arrival time of the traveling wave;

[0008] The edge node independently calculates the local fault location based on the fault feature vector extracted by the GPT model;

[0009] The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location.

[0010] A further improvement of the present invention is to detect the arrival time of traveling waves based on neural architecture search, specifically including:

[0011] The voltage signal of each node is measured by the sensor to preliminarily detect the arrival time of the traveling wave. The expression of the arrival time of the traveling wave is:

[0012] T i =min(t:|V i (t)-V ref (t)|>ε) (1)

[0013] Among them, T i is the arrival time of the traveling wave detected by the ith node, V i (t) is the voltage signal of the ith node, V ref (t) is the reference signal, ε is the preliminary detection threshold;

[0014] The measured voltage signal is preprocessed to extract the time series features. The expression is:

[0015] X i =Preprocess(V i (t)) (2)

[0016] Among them, X i is a time series feature, and Preprocess() is a data processing function;

[0017] Based on the time series characteristics, the neural architecture search algorithm is used to automatically find the best neural network architecture, which is expressed as:

[0018] N * (X i )=argminE(L(N(X i ),T i )) (3)

[0019] Among them, N * (X i) is the neural network structure that best fits the arrival time of the traveling wave, argmin represents the parameter of a function that obtains the minimum value in its domain, N(X i ) is the arrival time of the traveling wave predicted by the neural network, L(N(X i ),T i ) is the loss function, E() is the expected value;

[0020] Based on the optimal neural network architecture, the prediction of the arrival time of the traveling wave is expressed as:

[0021]

[0022] in, is the arrival time of the traveling wave predicted by the neural network;

[0023] The threshold of the traveling wave arrival time is dynamically adjusted according to the predicted traveling wave arrival time. The expression is:

[0024]

[0025] Among them, ε i is the traveling wave arrival time threshold, is the threshold adjustment function, θ is the parameter vector of the adjustment function;

[0026] The arrival time of the traveling wave is finally detected by using the arrival time threshold of the traveling wave. The expression is:

[0027] T i final =min(t:|V i (t)-V ref (t)|>ε i ) (6)

[0028] Among them, T i final is the final detected arrival time of the traveling wave, and t is the moment.

[0029] A further improvement of the present invention is to use the GPT model to analyze and model the historical data and the currently detected power grid data, and extract the fault feature vector, which specifically includes:

[0030] The propagation velocity v of the traveling wave and the detected arrival time of the traveling wave are used to perform preliminary fault location. The expression is:

[0031]

[0032] Among them, d i is the fault distance estimated based on the data of nodes i and j, v is the propagation speed of the traveling wave, T i , T jare the detected arrival times of the traveling waves at nodes i and j, respectively;

[0033] A microgrid fault feature prediction model based on the GPT model is constructed. The GPT model is used to analyze and model the historical data and the currently detected power grid data, and the fault feature vector is extracted. The expression is:

[0034] F i =GPT(e i ) (8)

[0035] e i =[v,T i final ,d i ,U i ,I i ] (9)

[0036] Among them: F i is the extracted fault feature vector, GPT(e i ) is the microgrid fault feature prediction model based on the GPT model, e i is the data sample input to the i-th node, U i is the voltage measured at node i, I i is the current measured at the i-th node.

[0037] A further improvement of the present invention is that the local fault location calculation performed by the edge node is:

[0038]

[0039] in: is the fault location calculated locally, LocalModel(F i ,T i ) is the local fault location model used by edge nodes, F i is the extracted fault feature vector, T i final is the final arrival time of the traveling wave detected by node i.

[0040] A further improvement of the present invention is to perform weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location, which specifically includes:

[0041] The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location, which is expressed as:

[0042]

[0043] Where: d final For the final microgrid fault location, wi is the weight of each edge node, is the fault location obtained by local calculation, and n is the number of edge nodes involved in the comprehensive calculation;

[0044] Perform confidence analysis on the final fault location location result to obtain the final fault location result, which is expressed as:

[0045]

[0046] Result=(d final ,Confidence) (13)

[0047] Among them: Confidence is the fault location confidence, is the variance of the estimated position of each edge node, is the fault location calculated locally, is the mean of the estimated positions of each edge node, and Result is the final fault location result.

[0048] The microgrid fault detection system based on traveling wave ranging and distributed edge computing of the present invention comprises:

[0049] A detection module is used to detect the arrival time of the traveling wave based on the neural architecture search, including: using sensors to measure the voltage signals of each node and extracting time series features; based on the time series features, using the neural architecture search algorithm to automatically find the best neural network architecture; based on the best neural network architecture, predicting the arrival time of the traveling wave, and dynamically adjusting the arrival time threshold of the traveling wave according to the predicted arrival time of the traveling wave; using the arrival time threshold of the traveling wave to finally detect the arrival time of the traveling wave;

[0050] A fault feature vector extraction module is used to analyze and model the historical data and the currently detected power grid data using the GPT model to extract the fault feature vector, wherein the currently detected power grid data includes the propagation speed of the traveling wave and the detected traveling wave arrival time;

[0051] The edge node calculation module is used to calculate the local fault location using the edge node based on the fault feature vector extracted by the GPT model;

[0052] The positioning module is used to perform weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location.

[0053] A further improvement of the present invention is that the specific operations performed by the detection module include:

[0054] The voltage signal of each node is measured by the sensor to preliminarily detect the arrival time of the traveling wave. The expression of the arrival time of the traveling wave is:

[0055] T i =min(t:|V i (t)-V ref (t)|>ε) (1)

[0056] Among them, T i is the arrival time of the traveling wave detected by the ith node, V i (t) is the voltage signal of the ith node, V ref (t) is the reference signal, ε is the preliminary detection threshold;

[0057] The measured voltage signal is preprocessed to extract the time series features. The expression is:

[0058] X i =Preprocess(V i (t)) (2)

[0059] Among them, X i is a time series feature, and Preprocess() is a data processing function;

[0060] Based on the time series characteristics, the neural architecture search algorithm is used to automatically find the best neural network architecture, which is expressed as:

[0061] N * (X i )=argminE(L(N(X i ),T i )) (3)

[0062] Among them, N * (X i ) is the neural network structure that best fits the arrival time of the traveling wave, argmin represents the parameter of a function that obtains the minimum value in its domain, N(X i ) is the arrival time of the traveling wave predicted by the neural network, L(N(X i ),T i ) is the loss function, E() is the expected value;

[0063] Based on the optimal neural network architecture, the prediction of the arrival time of the traveling wave is expressed as:

[0064]

[0065] in, is the arrival time of the traveling wave predicted by the neural network;

[0066] The threshold of the traveling wave arrival time is dynamically adjusted according to the predicted traveling wave arrival time. The expression is:

[0067]

[0068] Among them, ε i is the traveling wave arrival time threshold, is the threshold adjustment function, θ is the parameter vector of the adjustment function;

[0069] The arrival time of the traveling wave is finally detected by using the arrival time threshold of the traveling wave. The expression is:

[0070] T i final =min(t:|V i (t)-V ref (t)|>ε i ) (6)

[0071] Among them, T i final is the final detected arrival time of the traveling wave, and t is the moment.

[0072] A further improvement of the present invention is that the specific operations performed by the fault feature vector extraction module include:

[0073] The propagation velocity v of the traveling wave and the detected arrival time of the traveling wave are used to perform preliminary fault location. The expression is:

[0074]

[0075] Among them, d i is the fault distance estimated based on the data of nodes i and j, v is the propagation speed of the traveling wave, T i , T j are the detected arrival times of the traveling waves at nodes i and j, respectively;

[0076] A microgrid fault feature prediction model based on the GPT model is constructed. The GPT model is used to analyze and model the historical data and the currently detected power grid data, and the fault feature vector is extracted. The expression is:

[0077] F i =GPT(e i ) (8)

[0078] e i =[v,T i final ,d i ,U i ,I i ] (9)

[0079] Among them: F i is the extracted fault feature vector, GPT(e i ) is the microgrid fault feature prediction model based on the GPT model, ei is the data sample input to the i-th node, U i is the voltage measured at node i, I i is the current measured at the i-th node.

[0080] A further improvement of the present invention is that the edge node computing module performs the following calculation:

[0081]

[0082] in: is the fault location calculated locally, LocalModel(F i ,T i ) is the local fault location model used by edge nodes, F i is the extracted fault feature vector, T i final is the final arrival time of the traveling wave detected by node i.

[0083] A further improvement of the present invention is that the specific operations performed by the positioning module include:

[0084] The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location, which is expressed as:

[0085]

[0086] Where: d final For the final microgrid fault location, w i is the weight of each edge node, is the fault location obtained by local calculation, and n is the number of edge nodes involved in the comprehensive calculation;

[0087] Perform confidence analysis on the final fault location location result to obtain the final fault location result, which is expressed as:

[0088]

[0089] Result=(d final ,Confidence) (13)

[0090] Among them: Confidence is the fault location confidence, is the variance of the estimated position of each edge node, is the fault location calculated locally, is the mean of the estimated positions of each edge node, and Result is the final fault location result.

[0091] The beneficial effects of the present invention are as follows: the present invention can automatically design the neural network structure through NAS, find the optimal model structure to accurately predict the threshold of the arrival time of the traveling wave, and thus improve the accuracy of fault location. NAS can automatically adjust the network structure in the face of different noise and interference environments to enhance the robustness and adaptability of the model. The GPT model has a powerful sequence data processing capability, can extract high-level features from complex fault data, and improve the ability to identify various fault modes; through pre-training and fine-tuning, the present invention can adapt to different microgrid environments and fault types, and improve the generalization ability of the detection model in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 is a flow chart of a method in an embodiment of the present invention;

[0093] Figure 2 It is a schematic diagram of the system principle structure in an embodiment of the present invention. DETAILED DESCRIPTION

[0094] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0095] like Figure 1 As shown, this embodiment is a microgrid fault detection method based on traveling wave ranging and distributed edge computing, comprising the following steps:

[0096] Step 1, detecting the arrival time of the traveling wave based on the neural architecture search (NAS). It includes: using sensors to measure the voltage signals of each node, preliminarily detecting the arrival time of the traveling wave, preprocessing the measured voltage signals, and extracting time series features; using the neural architecture search algorithm (NAS) to automatically find the best neural network architecture to minimize the prediction error and find the neural network structure that can best fit the arrival time of the traveling wave; after finding the optimal neural network structure, using the neural architecture search algorithm (NAS) to predict the arrival time of the traveling wave, and dynamically adjust the arrival time threshold of the traveling wave according to the predicted arrival time of the traveling wave, and use the arrival time threshold of the traveling wave to finally detect the arrival time of the traveling wave; this embodiment can accurately predict the arrival time of the traveling wave by automatically designing the optimal neural network structure. Improve the accuracy and robustness of detection.

[0097] Step 2, use the GPT model to analyze and model the historical data and the currently detected power grid data, and extract the fault feature vector, where the currently detected power grid data includes the propagation speed of the traveling wave and the detected arrival time of the traveling wave. This embodiment first uses the propagation speed of the traveling wave and the detected arrival time of the traveling wave to perform preliminary fault location. Then construct a GPT fault prediction data sample vector, including the estimated fault distance, the propagation speed of the traveling wave, the detected arrival time of the traveling wave, the measured voltage, and the measured current. Finally, a microgrid fault feature prediction model based on the GPT model is constructed, and the data sample vector is input to achieve effective prediction of microgrid fault characteristics.

[0098] Step 3: The edge node independently calculates the local fault location based on the fault feature vector extracted by the GPT model and the detected traveling wave arrival time, thus locating the fault in the local area of ​​the microgrid.

[0099] Step 4: Perform weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location. Perform confidence analysis on the fault location result, and evaluate the confidence of the final fault location result based on the calculation results of each edge node to obtain the final fault location and confidence.

[0100] In step 1, the voltage signal of each node is measured by the sensor to preliminarily detect the arrival time of the traveling wave, as shown in expression (1):

[0101] T i =min(t:|V i (t)-V ref (t)|>ε) (1)

[0102] Among them, T i is the arrival time of the traveling wave detected by the ith node, V i (t) is the voltage signal of the ith node, V ref (t) is the reference signal, and ε is the preliminary detection threshold.

[0103] The measured voltage signal is preprocessed to extract the time series features. The expression is:

[0104] X i =Preprocess(V i (t)) (2)

[0105] Among them, X i It is a time series feature used as the input of the neural network, and Preprocess() is a data processing function.

[0106] Based on the time series characteristics, the neural architecture search algorithm is used to automatically find the best neural network architecture to minimize the prediction error. The goal is to find the neural network structure that can best fit the arrival time of the traveling wave, expressed as:

[0107] N * (X i )=argminE(L(N(X i ),T i )) (3)

[0108] Among them, N * (X i ) is the neural network structure that best fits the arrival time of the traveling wave, argmin represents the parameter of a function that obtains the minimum value in its domain, N(X i ) is the arrival time of the traveling wave predicted by the neural network, L(N(X i ),T i ) is the loss function, such as the mean square error (MSE), and E() is the expected value, which is usually estimated by the average loss on the training set.

[0109] Based on the optimal neural network architecture, the prediction of the arrival time of the traveling wave is expressed as:

[0110]

[0111] in, is the arrival time of the traveling wave predicted by the neural network;

[0112] According to the predicted arrival time of the traveling wave, the arrival time threshold of the traveling wave is dynamically adjusted, and the expression is:

[0113]

[0114] Among them, ε i is the traveling wave arrival time threshold, is the threshold adjustment function, which can be set according to specific business requirements, such as considering factors such as noise and device sensitivity, and θ is the parameter vector of the adjustment function;

[0115] The arrival time of the traveling wave is finally detected by using the arrival time threshold of the traveling wave. The expression is:

[0116] T i final =min(t:|V i (t)-V ref (t)|>ε i ) (6)

[0117] Among them, T i final is the final detected arrival time of the traveling wave, and t is the moment.

[0118] Step 2 specifically includes: using the propagation velocity v of the traveling wave and the detected arrival time of the traveling wave to perform preliminary fault location, the expression is:

[0119]

[0120] Among them, d i is the fault distance estimated based on the data of nodes i and j, v is the propagation speed of the traveling wave, T i , T j are the detected arrival times of the traveling waves at nodes i and j, respectively;

[0121] A microgrid fault feature prediction model based on the GPT model is constructed. The GPT model is used to analyze and model the historical data and the currently detected power grid data, and the fault feature vector is extracted. The expression is:

[0122] F i =GPT(e i ) (8)

[0123] e i =[v,T i final ,d i ,U i ,I i ] (9)

[0124] Among them: F i is the extracted fault feature vector, GPT(e i ) is the microgrid fault feature prediction model based on the GPT model, e i is the data sample input to the i-th node, U i is the voltage measured at node i, I i is the current measured at the i-th node.

[0125] The calculation performed by each edge node in step 3 is:

[0126]

[0127] in: is the fault location calculated locally, LocalModel(F i ,T i ) is the local fault location model used by edge nodes, F i is the extracted fault feature vector, T i final is the final arrival time of the traveling wave detected by node i.

[0128] Step 4 includes: performing weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location, which is expressed as:

[0129]

[0130] Where: d final For the final microgrid fault location, w i is the weight of each edge node, is the fault location obtained by local calculation, and n is the number of edge nodes involved in the comprehensive calculation;

[0131] Perform confidence analysis on the final fault location location result to obtain the final fault location result, which is expressed as:

[0132]

[0133] Result=(d final ,Confidence) (13)

[0134] Among them: Confidence is the fault location confidence, is the variance of the estimated position of each edge node, is the mean of the estimated positions of each edge node, and Result is the final fault location result.

[0135] The above method is implemented by the microgrid fault detection system based on traveling wave ranging and distributed edge computing in this embodiment. Figure 2 As shown, the system includes:

[0136] A detection module is used to detect the arrival time of the traveling wave based on the neural architecture search, including: using sensors to measure the voltage signals of each node and extracting time series features; based on the time series features, using the neural architecture search algorithm to automatically find the best neural network architecture; based on the best neural network architecture, predicting the arrival time of the traveling wave, and dynamically adjusting the arrival time threshold of the traveling wave according to the predicted arrival time of the traveling wave; using the arrival time threshold of the traveling wave to finally detect the arrival time of the traveling wave;

[0137] A fault feature vector extraction module is used to analyze and model the historical data and the currently detected power grid data using the GPT model to extract the fault feature vector, wherein the currently detected power grid data includes the propagation speed of the traveling wave and the detected traveling wave arrival time;

[0138] The edge node calculation module is used to calculate the local fault location using the edge node based on the fault feature vector extracted by the GPT model;

[0139] The positioning module is used to perform weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location.

[0140] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have the same meaning as in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined similarly here.

[0141] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A microgrid fault detection method based on traveling wave ranging and distributed edge computing, characterized in that: The following operations are included: Detecting the arrival time of traveling waves based on neural architecture search, including: using sensors to measure the voltage signals of each node and extracting time series features; using neural architecture search algorithms to automatically find the best neural network architecture based on the time series features; predicting the arrival time of traveling waves based on the best neural network architecture, and dynamically adjusting the threshold of the arrival time of traveling waves according to the predicted arrival time of traveling waves; using the threshold of the arrival time of traveling waves to finally detect the arrival time of traveling waves; The GPT model is used to analyze and model the historical data and the currently detected power grid data to extract the fault feature vector, wherein the currently detected power grid data includes the propagation speed of the traveling wave and the detected arrival time of the traveling wave; The edge node independently calculates the local fault location based on the fault feature vector extracted by the GPT model; The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location.

2. The microgrid fault detection method based on traveling wave ranging and distributed edge computing according to claim 1 is characterized in that: The detection of the arrival time of the traveling wave based on the neural architecture search specifically includes: The voltage signal of each node is measured by the sensor to preliminarily detect the arrival time of the traveling wave. The expression of the arrival time of the traveling wave is: T i =min(t:|V i (t)-V ref (t)|>ε) (1) Among them, T i is the arrival time of the traveling wave detected by the ith node, V i (t) is the voltage signal of the ith node, V ref (t) is the reference signal, ε is the preliminary detection threshold; The measured voltage signal is preprocessed to extract the time series features. The expression is: X i =Preprocess(V i (t)) (2) Among them, X i is a time series feature, and Preprocess() is a data processing function; Based on the time series characteristics, the neural architecture search algorithm is used to automatically find the best neural network architecture, which is expressed as: N * (X i )=argminE(L(N(X i ),T i )) (3) Among them, N * (X i ) is the neural network structure that best fits the arrival time of the traveling wave, argmin represents the parameter of a function that obtains the minimum value in its domain, N(X i ) is the arrival time of the traveling wave predicted by the neural network, L(N(X i ),T i ) is the loss function, E() is the expected value; Based on the optimal neural network architecture, the prediction of the arrival time of the traveling wave is expressed as: in, is the arrival time of the traveling wave predicted by the neural network; The threshold of the traveling wave arrival time is dynamically adjusted according to the predicted traveling wave arrival time. The expression is: Among them, ε i is the traveling wave arrival time threshold, is the threshold adjustment function, θ is the parameter vector of the adjustment function; The arrival time of the traveling wave is finally detected by using the arrival time threshold of the traveling wave. The expression is: T i final =min(t:|V i (t)-V ref (t)|>ε i ) (6) Among them, T i final is the final arrival time of the traveling wave detected by node i, and t is the time.

3. The microgrid fault detection method based on traveling wave ranging and distributed edge computing according to claim 1 is characterized in that: The GPT model is used to analyze and model the historical data and the currently detected power grid data to extract the fault feature vector, specifically including: The propagation velocity v of the traveling wave and the detected arrival time of the traveling wave are used to perform preliminary fault location. The expression is: Among them, d i is the fault distance estimated based on the data of nodes i and j, v is the propagation speed of the traveling wave, T i final , T j final are the arrival times of the traveling waves finally detected by nodes i and j respectively; A microgrid fault feature prediction model based on the GPT model is constructed. The GPT model is used to analyze and model the historical data and the currently detected power grid data, and the fault feature vector is extracted. The expression is: F i =GPT(e i ) (8) e i =[v,T i final ,d i ,U i ,I i ] (9) Among them: F i is the extracted fault feature vector, GPT(e i ) is the microgrid fault feature prediction model based on the GPT model, e i is the data sample input to the i-th node, U i is the voltage measured at node i, I i is the current measured at the i-th node.

4. The microgrid fault detection method based on traveling wave ranging and distributed edge computing according to claim 1 is characterized in that: The edge node calculates the local fault location as follows: in: is the fault location calculated locally, LocalModel(F i ,T i final ) is the local fault location model used by edge nodes, F i is the extracted fault feature vector, T i final is the final arrival time of the traveling wave detected by node i.

5. The microgrid fault detection method based on traveling wave ranging and distributed edge computing according to claim 1 is characterized in that: The weighted synthesis of the local fault locations calculated by all edge nodes to obtain the final fault location specifically includes: The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location, which is expressed as: Where: d final For the final microgrid fault location, w i is the weight of each edge node, is the fault location obtained by local calculation, and n is the number of edge nodes involved in the comprehensive calculation; Perform confidence analysis on the final fault location location result to obtain the final fault location result, which is expressed as: Among them: Confidence is the fault location confidence, is the variance of the estimated position of each edge node, is the fault location calculated locally, is the mean of the estimated positions of each edge node, and Result is the final fault location result.

6. Microgrid fault detection system based on traveling wave ranging and distributed edge computing, characterized by: The system comprises: A detection module is used to detect the arrival time of the traveling wave based on the neural architecture search, including: using sensors to measure the voltage signals of each node and extracting time series features; based on the time series features, using the neural architecture search algorithm to automatically find the best neural network architecture; based on the best neural network architecture, predicting the arrival time of the traveling wave, and dynamically adjusting the arrival time threshold of the traveling wave according to the predicted arrival time of the traveling wave; using the arrival time threshold of the traveling wave to finally detect the arrival time of the traveling wave; A fault feature vector extraction module is used to analyze and model the historical data and the currently detected power grid data using the GPT model to extract the fault feature vector, wherein the currently detected power grid data includes the propagation speed of the traveling wave and the detected traveling wave arrival time; The edge node calculation module is used to calculate the local fault location using the edge node based on the fault feature vector extracted by the GPT model; The positioning module is used to perform weighted synthesis on the local fault locations calculated by all edge nodes to obtain the final fault location.

7. The microgrid fault detection system based on traveling wave ranging and distributed edge computing according to claim 6 is characterized in that: The specific operations performed by the detection module include: The voltage signal of each node is measured by the sensor to preliminarily detect the arrival time of the traveling wave. The expression of the arrival time of the traveling wave is: T i =min(t:|V i (t)-V ref (t)|>ε) (1) Among them, T i is the arrival time of the traveling wave detected by the ith node, V i (t) is the voltage signal of the ith node, V ref (t) is the reference signal, ε is the preliminary detection threshold; The measured voltage signal is preprocessed to extract the time series features. The expression is: X i =Preprocess(V i (t)) (2) Among them, X i is a time series feature, and Preprocess() is a data processing function; Based on the time series characteristics, the neural architecture search algorithm is used to automatically find the best neural network architecture, which is expressed as: N * (X i )=argminE(L(N(X i ),T i )) (3) Among them, N * (X i ) is the neural network structure that best fits the arrival time of the traveling wave, argmin represents the parameter of a function that obtains the minimum value in its domain, N(X i ) is the arrival time of the traveling wave predicted by the neural network, L(N(X i ),T i ) is the loss function, E() is the expected value; Based on the optimal neural network architecture, the prediction of the arrival time of the traveling wave is expressed as: in, is the arrival time of the traveling wave predicted by the neural network; The threshold of the traveling wave arrival time is dynamically adjusted according to the predicted traveling wave arrival time. The expression is: Among them, ε i is the traveling wave arrival time threshold, is the threshold adjustment function, θ is the parameter vector of the adjustment function; The arrival time of the traveling wave is finally detected by using the arrival time threshold of the traveling wave. The expression is: T i final =min(t:|V i (t)-V ref (t)|>ε i ) (6) Among them, T i final is the final detected arrival time of the traveling wave, and t is the moment.

8. The microgrid fault detection system based on traveling wave ranging and distributed edge computing according to claim 6 is characterized in that: The specific operations performed by the fault feature vector extraction module include: The propagation velocity v of the traveling wave and the detected arrival time of the traveling wave are used to perform preliminary fault location. The expression is: Among them, d i is the fault distance estimated based on the data of nodes i and j, v is the propagation speed of the traveling wave, T i 、T j are the detected arrival times of the traveling waves at nodes i and j, respectively; A microgrid fault feature prediction model based on the GPT model is constructed. The GPT model is used to analyze and model the historical data and the currently detected power grid data, and the fault feature vector is extracted. The expression is: F i =GPT(e i ) (8) e i =[v,T i final ,d i ,U i ,I i ] (9) Among them: F i is the extracted fault feature vector, GPT(e i ) is the microgrid fault feature prediction model based on the GPT model, e i is the data sample input to the i-th node, U i is the voltage measured at node i, I i is the current measured at the i-th node.

9. The microgrid fault detection system based on traveling wave ranging and distributed edge computing according to claim 6, characterized in that: The calculation performed by the edge node calculation module is: in: is the fault location calculated locally, LocalModel(F i ,T i ) is the local fault location model used by edge nodes, F i is the extracted fault feature vector, T i final is the final arrival time of the traveling wave detected by node i.

10. The microgrid fault detection system based on traveling wave ranging and distributed edge computing according to claim 6, characterized in that: The specific operations performed by the positioning module include: The local fault locations calculated by all edge nodes are weighted and integrated to obtain the final fault location, which is expressed as: Where: d final For the final microgrid fault location, w i is the weight of each edge node, is the fault location obtained by local calculation, and n is the number of edge nodes involved in the comprehensive calculation; Perform confidence analysis on the final fault location location result to obtain the final fault location result, which is expressed as: Result=(d final ,Confidence) (13) Among them: Confidence is the fault location confidence, is the variance of the estimated position of each edge node, is the fault location calculated locally, is the mean of the estimated positions of each edge node, and Result is the final fault location result.

Images