Power grid information system fault positioning system based on adaptive learning

Abstract

The invention relates to the technical field of power grid information system fault positioning, and particularly discloses a power grid information system fault positioning system based on adaptive learning, and the system comprises a data collection module which is used for collecting electrical quantity data and state quantity data in the operation process of a power grid; the data preprocessing module is connected with the data acquisition module and is used for carrying out preprocessing operations such as filtering, denoising and normalization on the acquired data and removing interference and invalid information in the data; and the fault feature extraction module is connected with the data preprocessing module and is used for extracting feature parameters related to the power grid fault based on the preprocessed data. According to the power grid information system fault positioning system based on adaptive learning, a dynamic multi-scale adaptive learning algorithm is fused with reinforcement learning and transfer learning mechanisms, model parameters and structures are dynamically adjusted according to the real-time operation state of a power grid, similar scene experience is migrated, and the fault positioning accuracy and adaptability are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid information system fault location, and in particular to a power grid information system fault location system based on adaptive learning. Background Art

[0002] With the rapid development of social economy, the scale of power grids continues to expand and the structure becomes increasingly complex. The importance of safe and stable operation of power grids to social production and life becomes increasingly prominent. Once a power grid failure occurs, quickly and accurately locating the fault location is the key link to shorten the power outage time, reduce economic losses and ensure reliable power supply.

[0003] Traditional power grid fault location methods mainly include traveling wave method, impedance method and matrix algorithm based on switch quantity information. The traveling wave method determines the fault location by detecting the propagation time and speed of the fault traveling wave on the transmission line, but it is easily affected by traveling wave reflection, refraction and noise interference, resulting in reduced positioning accuracy; the impedance method calculates the fault location based on the relationship between the line impedance and voltage and current from the measurement point to the fault point, but the uncertainty of line parameters and changes in operating mode will seriously affect the positioning accuracy; the matrix algorithm based on switch quantity information relies on the action information of switches and protection devices. When the information is missing, erroneous, or there is refusal or malfunction, the reliability of the positioning result is difficult to guarantee. (

[0004] In recent years, with the development of artificial intelligence technology, adaptive learning algorithms have gradually been applied to the field of power grid fault location. Some existing solutions use traditional machine learning algorithms, such as neural networks and decision trees, to build fault location models, and achieve fault location by learning historical fault data. However, these algorithms have many limitations: on the one hand, once the model structure and parameters are determined, it is difficult to dynamically adjust them according to the real-time operating status of the power grid. When the power grid topology changes, the operating conditions change, or new fault types appear, the adaptability and accuracy of the model are significantly reduced; on the other hand, in terms of model evaluation, traditional methods often only focus on a single indicator of fault location accuracy, ignoring important factors such as positioning time, false alarm and omission rates, and model stability. It is impossible to comprehensively and accurately evaluate the model performance, and it is difficult to meet the power grid's requirements for high precision, high reliability, and strong adaptability of the fault location system. (

[0005] Therefore, there is an urgent need for a fault location system that can adapt to the complex operating environment of the power grid, dynamically adjust learning strategies, and adopt scientific and comprehensive evaluation methods to improve the accuracy and reliability of power grid fault location and ensure the safe and stable operation of the power grid. Summary of the Invention

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A power grid information system fault location system based on adaptive learning, comprising:

[0007] Data acquisition module, used to collect electrical quantity data and state quantity data during the operation of the power grid;

[0008] A data preprocessing module is connected to the data acquisition module and is used to perform preprocessing operations such as filtering, denoising, and normalization on the collected data to remove interference and invalid information in the data;

[0009] A fault feature extraction module is connected to the data preprocessing module and is used to extract characteristic parameters related to the power grid fault based on the preprocessed data, including fault current mutation characteristics, voltage phase change characteristics, and power direction characteristics;

[0010] An adaptive learning module, connected to the fault feature extraction module, has a built-in dynamic multi-scale adaptive learning algorithm. The algorithm integrates reinforcement learning and transfer learning mechanisms. It can dynamically adjust the learning strategy and transfer similar scenario knowledge based on changes in the power grid operating status and historical fault data, thereby achieving adaptive learning of power grid fault modes.

[0011] A fault location module, connected to the adaptive learning module, is used to input the extracted fault characteristic parameters into the algorithm model after adaptive learning to calculate the specific location of the power grid fault;

[0012] The communication module is used to transmit the fault location information obtained by the fault location module to the monitoring center or related operation and maintenance equipment.

[0013] Preferably, the dynamic multi-scale adaptive learning algorithm comprises the following steps:

[0014] Data initialization and scenario division: Obtain historical fault data and current grid operation data, and divide grid operation scenarios into different categories based on grid topology and operating load;

[0015] Reinforcement learning strategy construction: Define a state space containing fault characteristic parameters and model performance indicators, define an action space containing model parameter adjustment and structure optimization, and define a reward function based on improving fault location accuracy;

[0016] Transfer learning preprocessing: Calculate the similarity between new scenarios and historical scenarios, and select model parameters and feature extraction patterns for scenarios with high similarity;

[0017] Algorithmic iterative learning: Execute reinforcement learning actions at each time step, initialize the model with transfer learning knowledge, train and update the model using current scene data, store interaction experience in a replay pool, and regularly update the policy network.

[0018] Model evaluation and update: The model is evaluated using a weighted dynamic evaluation method that integrates multi-dimensional indicators. If the model meets the preset standards, it is applied to actual fault location; otherwise, iterative learning continues.

[0019] Preferably, the weighted dynamic evaluation method of integrating multi-dimensional indicators includes the following steps:

[0020] Construction of evaluation index system: Establish a multi-dimensional evaluation index system including fault location accuracy, location time, false alarm rate, missed alarm rate and model stability;

[0021] Dynamic weight determination: The weights of each evaluation indicator are adjusted in real time based on the grid operating status, including load level, operating topology, and weather conditions;

[0022] Data preparation: Select representative fault samples and normal operation samples from the test data set;

[0023] Model testing: Apply the model to be evaluated to process the test sample and record the positioning results and time;

[0024] Indicator calculation: Calculate the value of each evaluation indicator separately;

[0025] Weight application and comprehensive score calculation: Determine the weight of each indicator based on the current grid operation status, and calculate the comprehensive score of the model through weighted summation;

[0026] Result analysis and decision-making: Determine whether the model should be applied or optimized based on the comparison between the comprehensive score and the preset threshold.

[0027] Preferably, the algorithm model of the adaptive learning module includes but is not limited to a convolutional neural network model, a recurrent neural network model or a random forest model, and the model parameters and structure can be dynamically adjusted through reinforcement learning.

[0028] Preferably, the data acquisition module adopts a distributed sensor network, and the distributed sensor network is deployed at various nodes and key line locations of the power grid.

[0029] Preferably, the data preprocessing module uses a wavelet transform algorithm to filter the collected electrical quantity data and uses a normalization algorithm to normalize the data.

[0030] Preferably, it further includes a database module, which is connected to the data acquisition module, the fault feature extraction module and the adaptive learning module, and is used to store the collected original data, the extracted fault feature parameters, the historical fault data and the model training data.

[0031] Preferably, the adaptive learning module regularly retrains and optimizes the algorithm model based on the historical fault data and model training data stored in the database module.

[0032] Preferably, after calculating the specific location of the power grid fault, the fault location module is further used to evaluate the severity of the fault, and the evaluation method is based on the comparison of fault characteristic parameters with a preset severity threshold.

[0033] Preferably, the communication module uses a wireless network communication protocol or a power line carrier communication protocol for data transmission.

[0034] It has the following beneficial effects:

[0035] This power grid information system fault location system based on adaptive learning integrates reinforcement learning and transfer learning mechanisms through a dynamic multi-scale adaptive learning algorithm, dynamically adjusts model parameters and structure according to the real-time operating status of the power grid, transfers similar scenario experience, and significantly improves the accuracy and adaptability of fault location. It uses a reward function to drive model iteration and experience replay update, enhances the model's dynamic optimization capability, and avoids misjudgment in complex scenarios. It adopts a weighted dynamic evaluation method that integrates multi-dimensional indicators to flexibly adjust the evaluation focus according to the operating status of the power grid to achieve accurate evaluation. The distributed sensor network works in conjunction with data preprocessing and reinforcement learning to optimize the model structure, significantly shortening the fault location time. The introduction of model stability indicators and dynamic weight adjustment reduces the risk of false alarms and missed alarms. The database and adaptive learning module are linked to give the system the ability to continuously evolve, enabling it to automatically adapt to the trend of expanding power grid scale and complex structure, fully meeting the technical requirements of smart grids for autonomous evolution systems. DETAILED DESCRIPTION

[0036] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0037] The present invention provides a technical solution: a power grid information system fault location system based on adaptive learning, characterized by comprising:

[0038] Data acquisition module, used to collect electrical quantity data and state quantity data during the operation of the power grid;

[0039] A data preprocessing module is connected to the data acquisition module and is used to perform preprocessing operations such as filtering, denoising, and normalization on the collected data to remove interference and invalid information in the data;

[0040] A fault feature extraction module is connected to the data preprocessing module and is used to extract characteristic parameters related to the power grid fault based on the preprocessed data, including fault current mutation characteristics, voltage phase change characteristics, and power direction characteristics;

[0041] The adaptive learning module is connected to the fault feature extraction module and has a built-in dynamic multi-scale adaptive learning algorithm. The algorithm integrates reinforcement learning and transfer learning mechanisms. It can dynamically adjust the learning strategy and transfer similar scenario knowledge based on changes in the power grid operation status and historical fault data to achieve adaptive learning of power grid fault modes. The dynamic multi-scale adaptive learning algorithm includes the following steps:

[0042] Data initialization and scenario division: Obtain historical fault data and current grid operation data, and divide grid operation scenarios into different categories based on grid topology and operating load;

[0043] Reinforcement learning strategy construction: Define a state space containing fault characteristic parameters and model performance indicators, define an action space containing model parameter adjustment and structure optimization, and define a reward function based on improving fault location accuracy;

[0044] Transfer learning preprocessing: Calculate the similarity between new scenarios and historical scenarios, and select model parameters and feature extraction patterns for scenarios with high similarity;

[0045] Algorithmic iterative learning: Execute reinforcement learning actions at each time step, initialize the model with transfer learning knowledge, train and update the model using current scene data, store interaction experience in a replay pool, and regularly update the policy network.

[0046] Model evaluation and update: The model is evaluated using a weighted dynamic evaluation method that integrates multiple dimensions. If the model meets the preset criteria, it is applied to actual fault location. Otherwise, iterative learning continues. The weighted dynamic evaluation method integrates multiple dimensions includes the following steps:

[0047] Construction of evaluation index system: Establish a multi-dimensional evaluation index system including fault location accuracy, location time, false alarm rate, missed alarm rate and model stability;

[0048] Fault location accuracy P acc By formula Calculate, where N correct The number of times the model correctly locates the fault location, N total is the total number of fault test samples;

[0049] The positioning time T is the time interval from the model acquiring fault data to outputting the positioning result;

[0050] False alarm rate P false By formula Calculate, where N false-alarm N is the number of times the model misjudges normal samples as faults. normal is the total number of normal samples;

[0051] The false negative rate P miss By formula Calculate, where N miss-alarm is the number of times the model failed to detect the actual fault;

[0052] Model stability S is calculated by the standard deviation formula of multiple test results. Measure, where x i is the result of the i-th test, is the average value of multiple test results, and n is the number of tests;

[0053] Dynamic weight determination: The weights of the evaluation indicators are adjusted in real time based on the grid operation status, including load level, operation topology, and weather conditions. The dynamic weight determination steps include:

[0054] Establish a grid operation status assessment mechanism to monitor grid load levels, operating topology, weather conditions, and other factors in real time;

[0055] According to different grid operation states, dynamic weights are set for each evaluation indicator. A weight function is constructed, which takes the grid operation state parameters as input and outputs the weight values ​​of each indicator. The comprehensive score calculation formula is: Score = P acc ×w1+T×w2+P false ×w3+P miss ×w4+S×w5, where w1, w2, w3, w4, and w5 are the dynamic weights corresponding to fault location accuracy, location time, false alarm rate, missed alarm rate, and model stability, respectively;

[0056] Data preparation: Select representative fault samples and normal operation samples from the test data set;

[0057] Model testing: Apply the model to be evaluated to process the test sample and record the positioning results and time;

[0058] Indicator calculation: Calculate the value of each evaluation indicator separately;

[0059] Weight application and comprehensive score calculation: Determine the weight of each indicator based on the current grid operation status, and calculate the comprehensive score of the model through weighted summation;

[0060] Result analysis and decision-making: Determine whether to apply or optimize the model based on the comparison between the comprehensive score and the preset threshold;

[0061] A fault location module, connected to the adaptive learning module, is used to input the extracted fault characteristic parameters into the algorithm model after adaptive learning to calculate the specific location of the power grid fault;

[0062] A communication module, used to transmit the fault location information obtained by the fault location module to a monitoring center or related operation and maintenance equipment;

[0063] A database module is connected to the data acquisition module, the fault feature extraction module and the adaptive learning module, and is used to store the collected original data, the extracted fault feature parameters, the historical fault data and the model training data.

[0064] The working process of the power grid information system fault location system based on adaptive learning is as follows:

[0065] Data acquisition: The data acquisition module relies on a distributed sensor network to collect real-time electrical quantity data (such as voltage, current, and power) and status quantity data (such as switch status and protection action signals) during the operation of the power grid. These sensors are deployed at various nodes and key line locations of the power grid to ensure coverage of the entire range of power grid operation information;

[0066] Data preprocessing: The collected data is transmitted to the data preprocessing module, which uses the wavelet transform algorithm to filter the electrical quantity data to remove noise interference. The normalization algorithm is used to normalize the data, unify the data format and range, eliminate invalid information in the data, and provide a high-quality data foundation for subsequent analysis.

[0067] The pre-processed data enters the fault feature extraction module. This module extracts characteristic parameters related to the power grid fault based on electrical and state data. It focuses on fault current mutation characteristics (such as a sudden increase in current amplitude or an abnormal rate of change), voltage phase change characteristics (voltage phase offset before and after the fault), and power direction characteristics (change in power flow direction during the fault). These characteristic parameters will serve as the key basis for fault location.

[0068] Scenario division and data preparation: The adaptive learning module obtains historical fault data and current grid operation data. Based on factors such as grid topology and operating load, it divides grid operation scenarios into different categories, such as high-load operation scenarios, low-load operation scenarios, and special weather impact scenarios. Each scenario corresponds to a sub-dataset, providing targeted data for subsequent learning.

[0069] Reinforcement learning strategy execution: This defines a state space containing fault characteristic parameters and model performance indicators. Operations such as model parameter adjustment and structural optimization serve as the action space. A reward function is constructed based on improving fault location accuracy. At each time step, the algorithm selects an action from the action space based on the current grid operating scenario, such as adjusting the neural network's learning rate or changing the number of network layers.

[0070] Transfer learning knowledge application: Calculate the similarity between new scenarios and historical scenarios, screen out model parameters and feature extraction patterns for scenarios with high similarity, initialize or adjust parameters of the model after executing the action, and accelerate the model's learning process in the new scenario;

[0071] Model training and updating: The adjusted model is applied to the data of the current scenario for fault location training. The reward value is calculated based on the training results and fault location accuracy. The state, action, reward, next state and other information are stored in the experience replay pool. When the experience replay pool has reached a certain amount of data, data is randomly sampled for training, and the reinforcement learning policy network and value network are updated to continuously optimize the model parameters and structure.

[0072] Fault location: The fault location module inputs the extracted fault characteristic parameters into the adaptive learning algorithm model, calculates the specific location of the power grid fault through the model, and realizes accurate fault location.

[0073] Severity assessment: The fault location module compares fault characteristic parameters with preset severity thresholds to assess the severity of the fault, providing important reference for subsequent fault handling.

[0074] Data transmission: The communication module transmits the fault location information and severity assessment results obtained by the fault location module to the monitoring center or related operation and maintenance equipment through the wireless network communication protocol or power line carrier communication protocol, so that the operation and maintenance personnel can take timely measures;

[0075] Model evaluation and optimization: The model is evaluated using a weighted dynamic evaluation method that integrates multiple dimensions. First, a multi-dimensional evaluation indicator system is constructed, including fault location accuracy, location time, false alarm rate, missed alarm rate, and model stability. The weights of each indicator are adjusted in real time according to the grid operation status (load level, operating topology, weather conditions, etc.). An independent test data set is used for testing, and the values ​​of each indicator are calculated and weighted summed to obtain a comprehensive score. If the comprehensive score meets the preset standards, the model will continue to be used for actual fault location. If not, it will return to the adaptive learning module for retraining and optimization to ensure that the model performance always meets the grid operation requirements.

[0076] Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field and related fields without making creative efforts should fall within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described and explained in the present invention shall be implemented in accordance with conventional means in the field unless otherwise specified or limited.

Claims

1. A power grid information system fault location system based on adaptive learning, characterized in that: include: Data acquisition module, used to collect electrical quantity data and state quantity data during the operation of the power grid; A data preprocessing module is connected to the data acquisition module and is used to perform preprocessing operations such as filtering, denoising, and normalization on the collected data to remove interference and invalid information in the data; A fault feature extraction module is connected to the data preprocessing module and is used to extract characteristic parameters related to the power grid fault based on the preprocessed data, including fault current mutation characteristics, voltage phase change characteristics, and power direction characteristics; An adaptive learning module, connected to the fault feature extraction module, has a built-in dynamic multi-scale adaptive learning algorithm. The algorithm integrates reinforcement learning and transfer learning mechanisms. It can dynamically adjust the learning strategy and transfer similar scenario knowledge based on changes in the power grid operating status and historical fault data, thereby achieving adaptive learning of power grid fault modes. A fault location module, connected to the adaptive learning module, is used to input the extracted fault characteristic parameters into the algorithm model after adaptive learning to calculate the specific location of the power grid fault; The communication module is used to transmit the fault location information obtained by the fault location module to the monitoring center or related operation and maintenance equipment.

2. A power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: The dynamic multi-scale adaptive learning algorithm includes the following steps: Data initialization and scenario division: Obtain historical fault data and current grid operation data, and divide grid operation scenarios into different categories based on grid topology and operating load; Reinforcement learning strategy construction: Define a state space containing fault characteristic parameters and model performance indicators, define an action space containing model parameter adjustment and structure optimization, and define a reward function based on improving fault location accuracy; Transfer learning preprocessing: Calculate the similarity between new scenarios and historical scenarios, and select model parameters and feature extraction patterns for scenarios with high similarity; Algorithmic iterative learning: Execute reinforcement learning actions at each time step, initialize the model with transfer learning knowledge, train and update the model using current scene data, store interaction experience in a replay pool, and regularly update the policy network. Model evaluation and update: The model is evaluated using a weighted dynamic evaluation method that integrates multi-dimensional indicators. If the model meets the preset standards, it is applied to actual fault location; otherwise, iterative learning continues.

3. The adaptive learning-based power grid information system fault location system according to claim 2, characterized in that: The fusion multi-dimensional index weighted dynamic evaluation method includes the following steps: Construction of evaluation index system: Establish a multi-dimensional evaluation index system including fault location accuracy, location time, false alarm rate, missed alarm rate and model stability; Dynamic weight determination: The weights of each evaluation indicator are adjusted in real time based on the grid operating status, including load level, operating topology, and weather conditions; Data preparation: Select representative fault samples and normal operation samples from the test data set; Model testing: Apply the model to be evaluated to process the test sample and record the positioning results and time; Indicator calculation: Calculate the value of each evaluation indicator separately; Weight application and comprehensive score calculation: Determine the weight of each indicator based on the current grid operation status, and calculate the comprehensive score of the model through weighted summation; Result analysis and decision-making: Determine whether the model should be applied or optimized based on the comparison between the comprehensive score and the preset threshold.

4. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: The algorithm model of the adaptive learning module includes but is not limited to a convolutional neural network model, a recurrent neural network model or a random forest model, and the model parameters and structure can be dynamically adjusted through reinforcement learning.

5. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: The data acquisition module adopts a distributed sensor network, and the distributed sensor network is deployed at each node and key line position of the power grid.

6. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: The data preprocessing module uses a wavelet transform algorithm to filter the collected electrical quantity data and uses a normalization algorithm to normalize the data.

7. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: It also includes a database module, which is connected to the data acquisition module, the fault feature extraction module and the adaptive learning module, and is used to store the collected original data, the extracted fault feature parameters, the historical fault data and the model training data.

8. The power grid information system fault location system based on adaptive learning according to claim 7, characterized in that: The adaptive learning module regularly retrains and optimizes the algorithm model based on the historical fault data and model training data stored in the database module.

9. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: After calculating the specific location of the power grid fault, the fault location module is further used to evaluate the severity of the fault. The evaluation method is based on the comparison of fault characteristic parameters with a preset severity threshold.

10. The power grid information system fault location system based on adaptive learning according to claim 1, characterized in that: The communication module uses a wireless network communication protocol or a power line carrier communication protocol to transmit data.