Traveling wave distance measurement method and system based on machine learning

Abstract

The invention belongs to the technical field of power system fault location, and discloses a traveling wave fault location method and system based on machine learning. The method comprises the following steps: constructing a traveling wave signal enhancement model, a traveling wave distance measurement model and a fault maintenance strategy generation model; real-time power network topology data and real-time traveling wave data signals are collected and uploaded to a monitoring center; enhancing the real-time traveling wave data signal by using a traveling wave signal enhancement model to obtain an enhanced real-time traveling wave data signal; according to the real-time power network topology data and the enhanced real-time traveling wave data signal, using a traveling wave distance measurement model to carry out traveling wave distance measurement to obtain a real-time fault distance measurement result; and according to the real-time fault distance measurement result, generating a real-time fault maintenance strategy by using a fault maintenance strategy generation model, and publishing the real-time fault maintenance strategy to the power system. According to the invention, the problems of low ranging precision, difficulty in fault type identification and lack of response mechanism in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of power system fault location, and in particular relates to a traveling wave ranging method and system based on machine learning. Background Art

[0002] Power systems are essential infrastructure for modern society, and their safe and stable operation is crucial. Transmission line failures are one of the main factors affecting power system reliability. Rapidly and accurately locating fault locations (ranging) and developing effective repair strategies are crucial for shortening outages, reducing economic losses, and ensuring power supply reliability.

[0003] However, existing methods often face the following challenges: 1) Low ranging accuracy: Electromagnetic interference and background noise in the power system are superimposed on the traveling wave signal, making it difficult to accurately identify the arrival time and waveform characteristics of the initial traveling wave, resulting in increased ranging errors. Traveling waves are attenuated and distorted during transmission due to line losses, especially on long-distance or complex lines, making subsequent reflected and refracted waves difficult to distinguish from the initial wave. 2) Difficulty in fault type identification: Existing technologies rely solely on the characteristics of traveling wave signals to determine fault types (such as single-phase grounding and interphase short circuit). This leads to low accuracy when signal quality is poor or the fault situation is complex. 3) Lack of response mechanism: Existing technologies only implement traveling wave ranging to determine the fault location, but lack a subsequent response mechanism. In addition, the formulation of fault repair strategies with existing technologies relies heavily on the experience and judgment of dispatchers. This is not only inefficient but also easily influenced by personal experience and the mental state at the time, lacking consistency and objectivity. Summary of the Invention

[0004] In order to solve the problems of low ranging accuracy, difficulty in fault type identification and lack of response mechanism in the prior art, the present invention aims to provide a traveling wave ranging method and system based on machine learning.

[0005] The technical solution adopted in the present invention is: A traveling wave ranging method based on machine learning comprises the following steps: Using machine learning algorithms, a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model are built in the monitoring center. Collect the real-time power network topology data of the power system and the real-time traveling wave data signals of each transmission line, and upload them to the monitoring center; In the monitoring center, a traveling wave signal enhancement model is used to enhance the real-time traveling wave data signal to obtain an enhanced real-time traveling wave data signal; Based on the real-time power network topology data and the enhanced real-time traveling wave data signal, a traveling wave ranging model is used to perform traveling wave ranging and obtain real-time fault location results. According to the real-time fault location results, a fault repair strategy generation model is used to generate a strategy, obtain a real-time fault repair strategy, and publish it to the power system.

[0006] Furthermore, using machine learning algorithms, a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model are constructed in the monitoring center, including the following steps: In the monitoring center, a number of historical power network topology data and a number of historical traveling wave data signals are collected and preprocessed to obtain a number of preprocessed historical power network topology data and a number of preprocessed historical traveling wave data signals; Based on a number of pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and a number of enhanced historical traveling wave data signals are obtained; Based on several enhanced historical traveling wave data signals and several pre-processed historical power network topology data, a deep learning algorithm was used to construct a traveling wave ranging model and obtain several historical fault ranging results. Based on several historical fault location results and corresponding several historical power system operating states, a fault maintenance strategy generation model is constructed using a reinforcement learning algorithm.

[0007] Furthermore, the traveling wave signal enhancement model is constructed based on the PC-GAN algorithm, and the traveling wave signal enhancement model includes a generator constructed based on the CNN-LSTM algorithm and a discriminator constructed based on the CNN algorithm, which are connected in sequence. The generator is also provided with a power system physical constraint module.

[0008] Furthermore, based on the plurality of pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and a plurality of enhanced historical traveling wave data signals are obtained, including the following steps: Use the PC-GAN algorithm to build an initial traveling wave signal enhancement model; the initial traveling wave signal enhancement model includes an initial generator and an initial discriminator; Setting the generator loss function of the initial generator and the discriminator loss function of the initial discriminator, and setting the content loss function and the physical constraint dynamic penalty term of the power system physical constraint module; Based on the generator loss function, discriminator loss function, content loss function and physical constraint dynamic penalty term, the initial traveling wave signal enhancement model is alternately trained using several preprocessed historical traveling wave data signals to obtain the final traveling wave signal enhancement model and several enhanced historical traveling wave data signals.

[0009] Furthermore, the traveling wave ranging model is constructed based on the ST-GCN-Dyn-MLP algorithm, and the traveling wave ranging model includes a dynamic graph construction module constructed based on the Dynamic Graph algorithm, a spatiotemporal feature extraction module constructed based on the ST-GCN algorithm, and a ranging output module constructed based on the MLP algorithm, which are connected in sequence.

[0010] Furthermore, the fault repair strategy generation model is constructed based on the MARL-Grid algorithm, and the fault repair strategy generation model includes several intelligent agents and an environment simulator.

[0011] Furthermore, in the monitoring center, a traveling wave signal enhancement model is used to enhance the real-time traveling wave data signal to obtain an enhanced real-time traveling wave data signal, including the following steps: In the monitoring center, the real-time traveling wave data signal is preprocessed, and the obtained preprocessed real-time traveling wave data signal is input into the traveling wave signal enhancement model; Using a generator of a traveling wave signal enhancement model, extracting a first real-time spatiotemporal feature of the preprocessed real-time traveling wave data signal; The physical constraint module of the generator of the traveling wave signal enhancement model is used to constrain and calibrate the first real-time spatiotemporal feature, and output the enhanced real-time traveling wave data signal.

[0012] Furthermore, based on the real-time power network topology data and the enhanced real-time traveling wave data signal, a traveling wave ranging model is used to perform traveling wave ranging to obtain a real-time fault ranging result, including the following steps: Inputting real-time power network topology data, enhanced real-time traveling wave data signals of transmission lines and corresponding real-time power flow information into the traveling wave ranging model; Based on real-time power flow information, the dynamic graph construction module of the traveling wave ranging model is used to dynamically adjust the edge weights of the real-time adjacency matrix corresponding to the real-time power network topology data to obtain a real-time dynamic adjacency matrix; Using the spatiotemporal feature extraction module of the traveling wave ranging model, the second real-time spatiotemporal feature of the real-time dynamic adjacency matrix and the real-time signal time series corresponding to the enhanced real-time traveling wave data signal is extracted; According to the second real-time spatiotemporal feature, a ranging output module of a traveling wave ranging model is used to perform traveling wave ranging, and a real-time fault ranging result including a real-time fault type and a real-time fault distance is obtained.

[0013] Furthermore, based on the real-time fault location results, a fault repair strategy generation model is used to generate a strategy, obtain a real-time fault repair strategy, and publish it to the power system, including the following steps: Integrate the real-time fault location results and the real-time status monitoring data of the power system and encode them into a real-time status representation, and initialize the environment simulator and several intelligent agents based on the real-time status representation; Use the initialized environment simulator to convert the real-time state representation into real-time local observations for each initialized agent; Based on real-time local observations, the initialized agent is used to generate strategies and obtain the corresponding local real-time fault repair strategies. Combine the local real-time fault repair strategies of all initialized agents to obtain the initial real-time fault repair strategy; According to the initial real-time fault repair strategy, the internal state of the initialized environment simulator is updated to obtain an updated real-time state representation; Based on the updated real-time state representation, use the environment simulator to obtain the real-time reward of each initialized agent; Repeat the steps of updating the real-time state representation of the environment simulator and calculating the real-time reward of the agent, and output the optimal real-time fault repair strategy based on the accumulated real-time reward; According to the optimal real-time fault repair strategy, corresponding real-time executable instructions are generated and published to the power system.

[0014] A traveling wave ranging system based on machine learning is used to implement a traveling wave ranging method. The system is set up in a monitoring center and includes a data acquisition unit, a model construction unit, a traveling wave signal enhancement unit, a traveling wave ranging unit and a fault maintenance strategy generation unit connected in sequence. The monitoring center is communicated with several traveling wave acquisition devices of the power system respectively.

[0015] The beneficial effects of the present invention are: The present invention provides a traveling wave ranging method and system based on machine learning. Through the traveling wave signal enhancement model, it can effectively filter out noise and interference, extract purer and clearer traveling wave features, solve the problem of low ranging accuracy of traditional methods in strong noise environments, compensate and calibrate attenuated and distorted signals, and improve the ranging accuracy on long-distance lines or lines with complex structures; the traveling wave ranging model combines real-time power network topology data and enhanced traveling wave signals to better adapt to changes and unevenness of actual line parameters, extract real-time power network topology data, enhanced real-time traveling wave data signals of transmission lines, and accurately measure the distance between lines. and the corresponding real-time power flow information related to the fault type, to achieve real-time fault distance measurement result prediction including real-time fault type and real-time fault distance; the fault maintenance strategy generation model can automatically learn and optimize the maintenance strategy, avoiding the one-sidedness, subjectivity and inefficiency that may exist in manual experience-based decision-making. In a simulation environment, it learns how to comprehensively consider multiple factors such as fault location, type, maintenance resources (team, spare parts), real-time status of the power grid, and operation constraints to generate a globally optimal maintenance plan (such as optimal scheduling, resource allocation, and operation sequence), solving the problem that traditional methods are difficult to take into account multiple objectives and providing an intelligent response mechanism.

[0016] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of the traveling wave ranging method based on machine learning in the present invention.

[0018] Figure 2 It is a structural block diagram of the traveling wave ranging system based on machine learning in the present invention. DETAILED DESCRIPTION

[0019] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0020] Example 1: like Figure 1 As shown, this embodiment provides a traveling wave ranging method based on machine learning, including the following steps: S1: Using machine learning algorithms, a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model are constructed in the monitoring center. This includes the following steps: S1-1: In the monitoring center, a number of historical power network topology data, including but not limited to line connection relationships, line lengths, line parameters (resistance, inductance, capacitance), node information, switch status, etc., and a number of historical traveling wave data signals, including but not limited to position data, voltage traveling waves, current traveling waves, sampling value time series, phase information, frequency components, etc., are collected, and preprocessing including data cleaning, denoising (such as wavelet transform denoising), data synchronization, interpolation and completion is performed to obtain a number of preprocessed historical power network topology data and a number of preprocessed historical traveling wave data signals; S1-2: Based on a number of pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and a number of enhanced historical traveling wave data signals are obtained; The traveling wave signal enhancement model is built based on the Physics-Constrained Generative Adversarial Network (PC-GAN) algorithm. It consists of a generator based on the Convolutional Neural Network (CNN)-Long Short-Term Memory (LSTM) algorithm and a discriminator based on the CNN algorithm, connected in sequence. The generator also has a power system physical constraint module. This module constrains the generation process based on differential equations such as the telegraph equation and node power balance to ensure that the generated enhanced signal is physically reasonable and can better preserve fault characteristics. Based on several pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and several enhanced historical traveling wave data signals are obtained, including the following steps: S1-2-1: Use the PC-GAN algorithm to build an initial traveling wave signal enhancement model; the initial traveling wave signal enhancement model includes an initial generator and an initial discriminator; S1-2-2: Set the generator loss function of the initial generator and the discriminator loss function of the initial discriminator, and set the content loss function and the physical constraint dynamic penalty term of the power system physical constraint module; The formula is:

[0021]

[0022]

[0023]

[0024] Where, is the discriminator loss function; is the generator loss; is the input traveling wave data signal; is random noise; is the discriminator output function (probability between 0 and 1); Generates a function for enhanced signal; For all traveling wave data signals expectations; Traveling wave data signal The probability distribution of For all random noise expectations, is the probability distribution of random noise; is the L1 norm, which is used to calculate the sum of the absolute differences between two signals; is the content loss function; This penalty term is a physical constraint dynamic penalty term. It dynamically adjusts the weight based on the line parameters and operating status identified in real time, so that the generated signal is more consistent with the physical characteristics of the current line. is the dynamic weight; Line parameters and operating status; is a metric function that violates the laws of physics; S1-2-3: Based on the generator loss function, the discriminator loss function, the content loss function, and the physical constraint dynamic penalty term, the initial traveling wave signal enhancement model is alternately trained using a number of preprocessed historical traveling wave data signals to obtain the final traveling wave signal enhancement model and a number of enhanced historical traveling wave data signals; The formula is:

[0025]

[0026] Where, is the discriminator parameter gradient; is the discriminator parameter; is the total loss function of the discriminator; is the weight hyperparameter of content loss; is the learning rate; is the gradient of the total loss of the discriminator; For the generator parameters gradient; is the weight hyperparameter of the physical constraint loss; is the generator parameter; is the gradient of the total loss of the generator; S1-3: Based on several enhanced historical traveling wave data signals and several pre-processed historical power network topology data, a deep learning algorithm is used to construct a traveling wave ranging model and obtain several historical fault ranging results; The traveling wave ranging model is based on the Spatial-Temporal Graph Convolutional Network (ST-GCN), Dynamic Graph (Dyn), and Multi-Layer Perceptrons (MLP) algorithm. The model includes a sequentially connected dynamic graph construction module based on the Dynamic Graph algorithm, a spatiotemporal feature extraction module based on the ST-GCN algorithm, and a ranging output module based on the MLP algorithm. S1-4: Based on several historical fault location results and corresponding several historical power system operating states, a reinforcement learning algorithm is used to build a fault maintenance strategy generation model; The fault repair strategy generation model is built based on the Multi-Agent Reinforcement Learning in Grid Environments (MARL-Grid) algorithm, and the fault repair strategy generation model includes several agents and an environment simulator. S2: Collect the real-time power network topology data of the power system and the real-time traveling wave data signal of each transmission line, and upload them to the monitoring center; S3: In the monitoring center, the real-time traveling wave data signal is enhanced using the traveling wave signal enhancement model to obtain an enhanced real-time traveling wave data signal, including the following steps: S3-1: In the monitoring center, the real-time traveling wave data signal is pre-processed and the pre-processed real-time traveling wave data signal is obtained. Input traveling wave signal enhancement model, where is the time indicator; S3-2: using the generator of the traveling wave signal enhancement model, extracting the first real-time spatiotemporal features of the preprocessed real-time traveling wave data signal; The formula is:

[0027] Where, It is the first real-time spatiotemporal feature; It is a spatiotemporal feature extraction function consisting of a convolutional layer CNN and an LSTM layer; Real-time traveling wave data signal after preprocessing The converted pre-processed real-time traveling wave data signal vector; is the sampling indicator; S3-3: Using the physical constraint module of the generator of the traveling wave signal enhancement model, constraining and calibrating the first real-time spatiotemporal feature, and outputting an enhanced real-time traveling wave data signal, which has a higher signal-to-noise ratio and more complete fault feature information; The formula is:

[0028] Where, It is the first real-time spatiotemporal feature after constraint and calibration; Constraint and calibration functions for the physical constraint module; is the physical parameters of the line (such as line length, wave impedance, etc.); is the signal enhancement function; It is the enhanced real-time traveling wave data signal vector; S4: Based on the real-time power network topology data and the enhanced real-time traveling wave data signal, a traveling wave ranging model is used to perform traveling wave ranging to obtain a real-time fault ranging result, including the following steps: S4-1: Input the real-time power network topology data, the enhanced real-time traveling wave data signal of the transmission line, and the corresponding real-time power flow information into the traveling wave ranging model; S4-2: Based on real-time power flow information (active power on the line , reactive power , tidal current This information reflects the current load and electrical distance of the line, and can be summarized into a matrix ,in Indicates time line The comprehensive state characteristics on the real-time power network topology data can be, for example, normalized values ​​of power and current or combined characteristics), and the dynamic graph construction module of the traveling wave ranging model is used to dynamically adjust the real-time adjacency matrix corresponding to the real-time power network topology data. The edge weight , get the real-time dynamic adjacency matrix ; Represents a node in the power network and There is a line connection between If there is no direct connection, the following steps are included: S4-2-1: Using a fully connected network of dynamic graph building blocks to process input real-time power flow information , extract the key features of real-time traveling waves related to traveling wave propagation; for example, high-load lines may decay faster, and long-distance lines may take longer to propagate. Let the extracted features be ; S4-2-2: Using the gating mechanism of the dynamic graph building module to receive the real-time adjacency matrix and extracted real-time traveling wave key features As input, output is a Real-time weight matrices of the same shape The real-time weight matrix of this gate The computation of can be based on an attention mechanism or a simple weighted combination; S4-2-3: According to the real-time weight matrix To adjust the real-time adjacency matrix , get the dynamic adjacency matrix ; The formula is:

[0029] S4-3: Use the spatiotemporal feature extraction module of the traveling wave ranging model to extract the real-time dynamic adjacency matrix The real-time signal time series corresponding to the enhanced real-time traveling wave data signal The second real-time spatiotemporal feature reflecting the fault location and propagation characteristics includes the following steps: S4-3-1: Real-time dynamic adjacency matrix Standardize and obtain the standardized real-time dynamic adjacency matrix ; The formula is:

[0030] Where, To standardize the real-time dynamic adjacency matrix; for The diagonal matrix of S4-3-2: Use the fault propagation graph convolution layer of the spatiotemporal feature extraction module of the traveling wave ranging model to extract the real-time signal time series corresponding to the enhanced real-time traveling wave data signal. Perform graph convolution operation to obtain the real-time signal time series after graph convolution operation; The formula is:

[0031] Where, For time Node Real-time signal features after graph convolution operation; is a nonlinear activation function; Neighbor nodes Propagate to nodes The normalized real-time dynamic adjacency matrix of ; is the weight matrix; The node at the previous moment Real-time signal features after graph convolution operation; is the total number of neighbor nodes; is bias; S4-3-3: Use the one-dimensional convolution layer of the spatiotemporal feature extraction module of the traveling wave ranging model to recursively process the real-time signal time series after the graph convolution operation to obtain the final second real-time spatiotemporal feature ; S4-4: Based on the second real-time spatiotemporal feature , using the ranging output module of the traveling wave ranging model, traveling wave ranging is performed to obtain real-time fault ranging results including real-time fault type and real-time fault distance; S5: Based on the real-time fault location results, a fault repair strategy generation model is used to generate a strategy, obtain a real-time fault repair strategy, and publish it to the power system, including the following steps: S5-1: Integrate the real-time fault location results and the real-time status monitoring data of the power system and encode them into a real-time status representation. Initialize the environment simulator and several intelligent agents based on the real-time status representation. Real-time condition monitoring data includes: Load distribution: real-time load conditions of each node or area; Backup capacity: the currently available backup power generation capacity of the generator set and energy storage system; Weather information: current and recent weather conditions (e.g., rain and snow affecting traffic, high temperatures affecting equipment load, etc.); Network topology: The current connection status of the power grid (which lines / branches are connected and which are disconnected).

[0032] Historical information: recent faults, maintenance records, etc.; S5-2: Use the initialized environment simulator to convert the real-time state representation into the real-time local observation of each initialized agent = [fault location code, fault type code, node load vector, generator backup capacity vector, weather status code, current network topology matrix, available maintenance resource list, traffic condition information...]; S5-3: Based on the real-time local observation, use the initialized agent to generate a strategy and obtain the corresponding local real-time fault repair strategy; Agents include: : corresponds to the maintenance team A responsible for area R1; : corresponds to the maintenance team B responsible for area R2; : Corresponding to spare parts warehouse 1; : corresponds to the repair vehicle 1; : corresponds to the dispatch center operator (who may be responsible for higher-level decision-making or resource coordination); Action Space Repair team action space: = {move to location, start repairing fault, wait, request specific spare part}; Spare parts warehouse action space: = {Approve spare parts to maintenance team, reject request, notify estimated delivery time}; Repair car action space: = {Move to location to load spare parts and assist in repair}; Dispatch center action space: = {order to trip the line, order to close the disconnector, dispatch the maintenance team to the location}; Core objectives: Minimize fault isolation time: quickly isolate the fault area from the power grid to reduce the impact on non-fault areas; Minimize system recovery time: quickly repair faults and restore power; Minimize resource consumption: rationally use manpower, material resources and time; Minimize the impact on users: reduce the number of users with power outages and the duration of power outages; S5-4: Combine the local real-time fault repair strategies of all initialized agents to obtain the initial real-time fault repair strategy; Initial real-time fault repair strategy: Maintenance teams: move to a point, start maintenance operations, request specific spare parts, and report progress; Spare parts warehouse: approve / reject spare parts requests and arrange spare parts transportation; Dispatch center: issues coordination instructions; S5-5: updating the internal state of the initialized environment simulator according to the initial real-time fault repair strategy to obtain an updated real-time state representation; S5-6: Based on the updated real-time state representation, use the environment simulator to obtain the real-time reward of each initialized agent; Real-time rewards need to balance multiple objectives (time, cost, recovery speed, user impact, etc.) and include: Periodic rewards (such as successfully isolating a fault or completing key repair steps); Penalty items (such as timeout, resource waste, and user impact); Terminal rewards (such as full restoration of power); Cooperation rewards (encourage agents to work together); S5-7: Repeat the steps of updating the real-time state representation of the environment simulator and calculating the real-time reward of the agent, and perform a large number of interactions in the simulated environment until the agent's strategy converges to a level that can effectively solve the fault repair problem. Based on the accumulated real-time reward, the optimal real-time fault repair strategy is output; The optimal real-time fault repair strategy includes: recommended tripping range, optimal maintenance team scheduling plan, spare parts deployment plan, fault isolation and recovery operation sequence; S5-8: Generate corresponding real-time executable instructions according to the optimal real-time fault repair strategy, and publish the real-time executable instructions to the power system.

[0033] Example 2: like Figure 2 As shown, this embodiment provides a traveling wave ranging system based on machine learning, which is used to implement a traveling wave ranging method. The system is arranged in a monitoring center, and the system includes a data acquisition unit, a model building unit, a traveling wave signal enhancement unit, a traveling wave ranging unit and a fault repair strategy generation unit connected in sequence. The monitoring center is respectively communicated with several traveling wave acquisition devices of the power system.

[0034] Traveling wave acquisition devices are installed on both sides of the transmission line to collect real-time traveling wave data signals of each transmission line in the power system and upload them to the monitoring center; A data acquisition unit is used to receive the real-time traveling wave data signal of each transmission line in the power system uploaded by the traveling wave acquisition device and the real-time power network topology data of the power system uploaded by the power system server; A model building unit, used to use machine learning algorithms to build a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model in the monitoring center; A traveling wave signal enhancement unit is used to enhance the real-time traveling wave data signal in the monitoring center using a traveling wave signal enhancement model to obtain an enhanced real-time traveling wave data signal; A traveling wave ranging unit is used to perform traveling wave ranging based on real-time power network topology data and enhanced real-time traveling wave data signals using a traveling wave ranging model to obtain real-time fault location results; The fault repair strategy generation unit is used to generate a strategy based on the real-time fault location result using a fault repair strategy generation model, obtain a real-time fault repair strategy, and publish it to the power system.

[0035] The present invention provides a traveling wave ranging method and system based on machine learning. Through the traveling wave signal enhancement model, it can effectively filter out noise and interference, extract purer and clearer traveling wave features, solve the problem of low ranging accuracy of traditional methods in strong noise environments, compensate and calibrate attenuated and distorted signals, and improve the ranging accuracy on long-distance lines or lines with complex structures; the traveling wave ranging model combines real-time power network topology data and enhanced traveling wave signals to better adapt to changes and unevenness of actual line parameters, extract real-time power network topology data, enhanced real-time traveling wave data signals of transmission lines, and accurately measure the distance between lines. and the corresponding real-time power flow information related to the fault type, to achieve real-time fault distance measurement result prediction including real-time fault type and real-time fault distance; the fault maintenance strategy generation model can automatically learn and optimize the maintenance strategy, avoiding the one-sidedness, subjectivity and inefficiency that may exist in manual experience-based decision-making. In a simulation environment, it learns how to comprehensively consider multiple factors such as fault location, type, maintenance resources (team, spare parts), real-time status of the power grid, and operation constraints to generate a globally optimal maintenance plan (such as optimal scheduling, resource allocation, and operation sequence), solving the problem that traditional methods are difficult to take into account multiple objectives and providing an intelligent response mechanism.

[0036] The present invention is not limited to the above optional embodiments. Anyone can derive various other forms of products based on the teachings of the present invention. The above specific embodiments should not be construed as limiting the scope of protection of the present invention. The scope of protection of the present invention shall be based on the scope defined in the claims, and the description can be used to interpret the claims.

Claims

1. A traveling wave ranging method based on machine learning, characterized by: The steps include: Using machine learning algorithms, a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model are built in the monitoring center. Collect the real-time power network topology data of the power system and the real-time traveling wave data signals of each transmission line, and upload them to the monitoring center; In the monitoring center, a traveling wave signal enhancement model is used to enhance the real-time traveling wave data signal to obtain an enhanced real-time traveling wave data signal; Based on the real-time power network topology data and the enhanced real-time traveling wave data signal, a traveling wave ranging model is used to perform traveling wave ranging and obtain real-time fault location results. According to the real-time fault location results, a fault repair strategy generation model is used to generate a strategy, obtain a real-time fault repair strategy, and publish it to the power system.

2. The traveling wave ranging method based on machine learning according to claim 1, characterized in that: Using machine learning algorithms, a traveling wave signal enhancement model, a traveling wave ranging model, and a fault repair strategy generation model are constructed in the monitoring center. The steps include: In the monitoring center, a number of historical power network topology data and a number of historical traveling wave data signals are collected and preprocessed to obtain a number of preprocessed historical power network topology data and a number of preprocessed historical traveling wave data signals; Based on a number of pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and a number of enhanced historical traveling wave data signals are obtained; Based on several enhanced historical traveling wave data signals and several pre-processed historical power network topology data, a deep learning algorithm was used to construct a traveling wave ranging model and obtain several historical fault ranging results. Based on several historical fault location results and corresponding several historical power system operating states, a fault maintenance strategy generation model is constructed using a reinforcement learning algorithm.

3. The traveling wave ranging method based on machine learning according to claim 2, characterized in that: The traveling wave signal enhancement model is constructed based on the PC-GAN algorithm, and the traveling wave signal enhancement model includes a generator constructed based on the CNN-LSTM algorithm and a discriminator constructed based on the CNN algorithm, which are connected in sequence. The generator is also provided with a power system physical constraint module.

4. The traveling wave ranging method based on machine learning according to claim 3, characterized in that: Based on several pre-processed historical traveling wave data signals, a deep learning algorithm is used to construct a traveling wave signal enhancement model, and several enhanced historical traveling wave data signals are obtained, including the following steps: Using the PC-GAN algorithm, an initial traveling wave signal enhancement model is constructed; the initial traveling wave signal enhancement model includes an initial generator and an initial discriminator; Setting the generator loss function of the initial generator and the discriminator loss function of the initial discriminator, and setting the content loss function and the physical constraint dynamic penalty term of the power system physical constraint module; Based on the generator loss function, discriminator loss function, content loss function and physical constraint dynamic penalty term, the initial traveling wave signal enhancement model is alternately trained using several preprocessed historical traveling wave data signals to obtain the final traveling wave signal enhancement model and several enhanced historical traveling wave data signals.

5. The traveling wave ranging method based on machine learning according to claim 4, characterized in that: The traveling wave ranging model is constructed based on the ST-GCN-Dyn-MLP algorithm, and the traveling wave ranging model includes a dynamic graph construction module constructed based on the DynamicGraph algorithm, a spatiotemporal feature extraction module constructed based on the ST-GCN algorithm, and a ranging output module constructed based on the MLP algorithm, which are connected in sequence.

6. The traveling wave ranging method based on machine learning according to claim 5, characterized in that: The fault repair strategy generation model is constructed based on the MARL-Grid algorithm, and the fault repair strategy generation model includes several intelligent agents and an environment simulator.

7. The traveling wave ranging method based on machine learning according to claim 6, characterized in that: In the monitoring center, a traveling wave signal enhancement model is used to enhance the real-time traveling wave data signal to obtain an enhanced real-time traveling wave data signal, including the following steps: In the monitoring center, the real-time traveling wave data signal is preprocessed, and the obtained preprocessed real-time traveling wave data signal is input into the traveling wave signal enhancement model; Using a generator of a traveling wave signal enhancement model, extracting a first real-time spatiotemporal feature of the preprocessed real-time traveling wave data signal; The physical constraint module of the generator of the traveling wave signal enhancement model is used to constrain and calibrate the first real-time spatiotemporal feature, and output the enhanced real-time traveling wave data signal.

8. The traveling wave ranging method based on machine learning according to claim 7, characterized in that: Based on the real-time power network topology data and the enhanced real-time traveling wave data signal, a traveling wave ranging model is used to perform traveling wave ranging to obtain real-time fault location results, including the following steps: Inputting real-time power network topology data, enhanced real-time traveling wave data signals of transmission lines and corresponding real-time power flow information into the traveling wave ranging model; Based on real-time power flow information, the dynamic graph construction module of the traveling wave ranging model is used to dynamically adjust the edge weights of the real-time adjacency matrix corresponding to the real-time power network topology data to obtain a real-time dynamic adjacency matrix; Using the spatiotemporal feature extraction module of the traveling wave ranging model, the second real-time spatiotemporal feature of the real-time dynamic adjacency matrix and the real-time signal time series corresponding to the enhanced real-time traveling wave data signal is extracted; According to the second real-time spatiotemporal feature, a ranging output module of a traveling wave ranging model is used to perform traveling wave ranging, and a real-time fault ranging result including a real-time fault type and a real-time fault distance is obtained.

9. The traveling wave ranging method based on machine learning according to claim 8, characterized in that: Based on the real-time fault location results, a fault repair strategy generation model is used to generate a strategy, obtain a real-time fault repair strategy, and publish it to the power system. The process includes the following steps: Integrate the real-time fault location results and the real-time status monitoring data of the power system and encode them into a real-time status representation, and initialize the environment simulator and several intelligent agents based on the real-time status representation; Use the initialized environment simulator to convert the real-time state representation into real-time local observations for each initialized agent; Based on real-time local observations, the initialized agent is used to generate strategies and obtain the corresponding local real-time fault repair strategies. Combine the local real-time fault repair strategies of all initialized agents to obtain the initial real-time fault repair strategy; According to the initial real-time fault repair strategy, the internal state of the initialized environment simulator is updated to obtain an updated real-time state representation; Based on the updated real-time state representation, use the environment simulator to obtain the real-time reward of each initialized agent; Repeat the steps of updating the real-time state representation of the environment simulator and calculating the real-time reward of the agent, and output the optimal real-time fault repair strategy based on the accumulated real-time reward; According to the optimal real-time fault repair strategy, corresponding real-time executable instructions are generated and published to the power system.

10. A traveling wave ranging system based on machine learning, used to implement the traveling wave ranging method according to any one of claims 1 to 9, characterized in that: The system is set up in a monitoring center, and the system includes a data acquisition unit, a model building unit, a traveling wave signal enhancement unit, a traveling wave ranging unit and a fault maintenance strategy generation unit connected in sequence. The monitoring center is respectively communicated with several traveling wave acquisition devices of the power system.

Images