A transmission line fault early warning system and method based on big data drive

Abstract

The present invention relates to the technical field of electric line fault detection, and discloses a transmission line fault warning system and method based on big data drive, the system comprising a state prediction module, a risk assessment module, a warning decision module and a management execution module. The state prediction module uses historical operation data, environmental conditions and line parameters to build a prediction model and output future state indicators; the risk assessment module comprehensively considers multi-source information such as real-time operation data and historical maintenance records to evaluate the current fault risk index; the warning decision module models the management process as a Markov decision process, and makes warning decisions based on the outputs of the first two modules; the management execution module is responsible for real-time management and execution of decision results. The present invention improves the accuracy and real-time performance of transmission line fault warnings by comprehensively utilizing big data technology, overcomes the shortcomings of traditional fault management methods, and provides a strong guarantee for the safe and stable operation of the power system.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line fault detection, and in particular to a power transmission line fault early warning system and method driven by big data. Background Art

[0002] With the rapid development of the power system and the continuous expansion of the power grid, the safe and stable operation of transmission lines, as an important channel for power transmission, plays a vital role in ensuring power supply and maintaining social and economic order. However, due to exposure to the natural environment, transmission lines are susceptible to climate change, equipment aging, human damage and other factors, and the risk of failure always exists. Traditional transmission line fault management mainly relies on regular inspections and post-fault repairs. These methods are not only inefficient, but also difficult to detect potential faults in a timely and accurate manner. They often lead to measures being taken only after the fault occurs, which brings instability to the power supply. With the rapid development and widespread application of big data technology, new ideas and methods have been provided for the fault management of transmission lines. Big data technology can process and analyze massive, multi-source, heterogeneous data, dig out the laws and patterns hidden behind the data, and provide strong support for the status monitoring, risk assessment and early warning decision-making of transmission lines.

[0003] The existing transmission line fault warning systems and methods still have some shortcomings. The state prediction model often only considers a single or a few factors, ignoring the complexity and diversity of the transmission line operation status, resulting in inaccurate prediction results. The risk assessment model lacks comprehensive consideration of multi-source information such as real-time operation data and historical maintenance records, making it difficult to fully reflect the fault risk status of the transmission line. The warning decision-making process often relies on manual experience and rules, lacks scientificity and systematicity, and is difficult to meet the requirements of modern power grids for real-time and accuracy of fault warning. Summary of the invention

[0004] The purpose of the present invention is to provide a transmission line fault early warning system and method driven by big data to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a transmission line fault early warning system driven by big data, the system comprising:

[0006] The state prediction module is used to build and run a transmission line state prediction model, which receives historical operation data, environmental conditions and line parameters of the transmission line as input, and outputs future state indicator data of the transmission line in the future period of time; when training the model, the historical operation data, historical environmental condition data, line parameters and corresponding actual operation state data are used as training data;

[0007] The risk assessment module is used to build and run a fault risk assessment model, which receives the real-time operation data, historical maintenance records, line parameters and environmental condition data of the transmission line as input, and outputs the current fault risk index of the transmission line; when training the model, the historical operation data, maintenance records, line parameters, environmental condition data and corresponding fault records are used as training data;

[0008] The early warning decision module is used to build and run the fault early warning decision model, which models the management process of the transmission line as a Markov decision process and makes fault early warning decisions based on the output of the state prediction module and the output of the risk assessment module;

[0009] The management execution module is used to perform real-time management based on the trained status prediction module, risk assessment module and early warning decision module, and automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system.

[0010] Preferably, the specific definition of the fault warning decision model includes:

[0011] State space: defined as the combination of real-time operation data of the transmission line, line parameters and environmental condition data;

[0012] Action space: defined as the set of executable inspection, maintenance, adjustment and protection operations;

[0013] Reward function: Calculate rewards or penalties based on state prediction deviation and risk index deviation, give positive rewards to operations that reduce deviations, and give penalties to operations that increase deviations;

[0014] Output strategy: It is defined as the method of selecting the best action in the current state, and the decision actions generated are continuously optimized through deep reinforcement learning algorithms.

[0015] Preferably, the implementation method of the reward function includes:

[0016] Obtain real-time operation status data of the transmission line; compare the real-time operation status data with the predicted status indicator data, and calculate the status prediction deviation;

[0017] Obtain real-time fault risk assessment data of the transmission line; compare the current fault risk index output by the fault risk assessment model with the preset risk standard or historical risk data, and calculate the risk index deviation;

[0018] Rewards or penalties are calculated based on the state prediction deviation and risk index deviation. The specific formula is:

[0019] Reward = - (α * state prediction deviation + β * risk index deviation), where α and β are the weight coefficients of the state prediction deviation and risk index deviation reward items respectively, and α, β > 0.

[0020] Preferably, the real-time operation data includes current, voltage, power factor, temperature and line impedance; the environmental condition data includes ambient temperature, humidity, wind speed, wind direction and lightning activity; the line parameters include line length, material, cross-sectional area and installation height; the maintenance record includes maintenance time, maintenance type, maintenance cost and maintenance effect;

[0021] The executable inspection, maintenance, adjustment and protection operation set includes: line inspection, component replacement, tightening connections, adjusting line tension, installing lightning protection equipment and adjusting protection device settings.

[0022] Preferably, the transmission line state prediction model is constructed using a convolutional long short-term memory network ConvLSTM algorithm, and its network structure includes:

[0023] Input layer: used to receive training data of the model;

[0024] Hidden layer: contains at least one ConvLSTM hidden layer. Each ConvLSTM hidden layer contains several ConvLSTM units. Each ConvLSTM unit has a forget gate, input gate and output gate to control the flow of information and state update, while considering spatial features.

[0025] Output layer: Receives the output of the hidden layer and maps the output of the ConvLSTM network to the predicted operating status indicator value through a fully connected layer.

[0026] Preferably, the step of training the transmission line state prediction model comprises:

[0027] F101: Initialize the weights and biases of the ConvLSTM network using a random initialization method;

[0028] F102: Select the loss function to measure the difference between the model prediction value and the true value. The mean square error MSE is used as the loss function.

[0029] F103: Use the back propagation algorithm and Adam optimization algorithm to calculate the gradient according to the loss function, update the weights and bias items of the ConvLSTM network, and iteratively train the network until the loss function converges;

[0030] F104: Save the trained ConvLSTM network model.

[0031] Preferably, the fault risk assessment model is constructed using a random forest RF algorithm, and its training steps include:

[0032] F201: Use the historical operation data, maintenance records, line parameters and environmental condition data in the training data as input features and the corresponding fault records as labels to construct a training sample set for the random forest algorithm;

[0033] F202: Use the random forest algorithm to train the training sample set, and improve the accuracy and robustness of the model by building multiple decision trees and combining their results;

[0034] F203: During the training process, the number and maximum depth of trees of the random forest algorithm are optimized by cross-validation.

[0035] Preferably, the fault warning decision model is constructed using a proximal strategy optimization (PPO) algorithm, and its training steps include:

[0036] S1. Initialize the policy network and value network: the policy network is used to generate action probability distribution, and the value network is used to estimate state value;

[0037] S2. Initialize the experience playback buffer D: set the buffer size N to store the state transition experience generated during the power transmission line management process, including state s, action a, reward r, next state s' and whether the termination flag is done;

[0038] S3. For each training round:

[0039] S301: Initialize the state s at the beginning of the round;

[0040] S302: Before the round is terminated, the following steps are executed in a loop: i) the policy network selects action a according to the current state s and adds exploration noise; ii) executes action a, observes the reward r, the next state s' and the termination flag done; iii) stores the experience (s, a, r, s', done) in the experience replay buffer D; iv) randomly samples a batch of experience from the experience replay buffer D to update the network parameters; v) uses the sampled experience to update the parameters of the value network to minimize the prediction error of the value function; vi) uses the sampled experience to update the parameters of the policy network and optimizes it through the objective function of the PPO algorithm; vii) updates the current state s to the next state s';

[0041] S303: When the round ends, reset the state s to the new round start state;

[0042] S4: During the training process, the performance of the fault warning decision model is regularly evaluated, and based on the evaluation results, the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise, and the size of the experience replay buffer are adjusted;

[0043] S5: Repeat steps S3 and S4 until the model performance reaches a preset standard.

[0044] Preferably, in step S4, the accumulated reward value is used as the performance evaluation indicator of the current round of training, and the specific implementation steps include:

[0045] S401: After each training round, the cumulative reward value of the round is calculated as the performance evaluation indicator of the current round iteration;

[0046] S402: If the cumulative reward value of the current iteration is higher than the cumulative reward value of the previous iteration, the current model parameters are kept unchanged and the next iteration is continued;

[0047] S403: If the cumulative reward value of the current round of iterations is lower than the cumulative reward value of the previous round of iterations, adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise and the size of the experience replay buffer, reinitialize the parameters of the policy network and the value network, and start a new round of training.

[0048] Preferably, a transmission line fault early warning method based on big data drive comprises:

[0049] F1: Construct a transmission line state prediction model to predict the operation state of the transmission line in the future. Its input data includes: historical operation data, environmental conditions and line parameters of the transmission line. Its output data is: future state indicator data of the transmission line. Model training data includes: historical operation data, historical environmental conditions data, line parameters and corresponding actual operation state data.

[0050] F2: Construct a fault risk assessment model to assess the current fault risk of the transmission line; its input data includes: real-time operation data of the transmission line, historical maintenance records, line parameters and environmental condition data; its output data is: the current fault risk index of the transmission line; model training data includes: historical operation data, maintenance records, line parameters, environmental condition data and corresponding fault records;

[0051] F3: Construct a fault warning decision model, model the management process of the transmission line as a Markov decision process, and make fault warning decisions based on the output of the transmission line state prediction model and the output of the fault risk assessment model;

[0052] F4: Real-time management is carried out based on the trained transmission line status prediction model, fault risk assessment model and fault warning decision model, and the decision results are automatically executed or transmitted to the management personnel for confirmation and execution through the management system.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] The present invention constructs and runs a transmission line state prediction model, taking into account multiple sources of information such as historical operation data of the transmission line, environmental conditions, and line parameters. This comprehensive data fusion method significantly improves the accuracy of state prediction, can more accurately reflect the operation state of the transmission line in the future, and provides a reliable basis for subsequent fault risk assessment and early warning decision-making.

[0055] The risk assessment module builds a comprehensive fault risk assessment model by receiving the real-time operation data, historical maintenance records, line parameters and environmental condition data of the transmission line. The model can comprehensively consider the impact of various factors on the fault risk of the transmission line, accurately output the current fault risk index, and provide managers with intuitive fault risk assessment results, which helps to take preventive measures in time. The early warning decision module models the management process of the transmission line as a Markov decision process, and makes fault early warning decisions based on the outputs of the state prediction module and the risk assessment module. This scientific decision-making method not only improves the real-time nature of the early warning, but also ensures the scientific and systematic nature of the decision. Once a potential fault is predicted or a high fault risk is assessed, the system can immediately issue an early warning, which wins valuable time for fault handling.

[0056] The management execution module can perform real-time management based on the trained state prediction module, risk assessment module and early warning decision module, and automatically execute the decision results or transmit them to the management personnel for confirmation through the management system. This automated management method greatly reduces manual intervention, improves management efficiency, and reduces misjudgments and delays caused by human factors.

[0057] Through the implementation of the present invention, the fault warning and management of the transmission line become more accurate, real-time and efficient. This helps to timely discover and deal with potential faults, reduce the possibility of faults, thereby ensuring the safe and stable operation of the power system, maintaining social and economic order, and improving the reliability and stability of power supply. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 This is a working principle diagram of a power transmission line fault early warning method based on big data drive according to the present invention;

[0059] Figure 2 Flowchart for training a transmission line state prediction model;

[0060] Figure 3 Flowchart of the fault warning decision model constructed by using the proximal strategy to optimize the PPO algorithm. DETAILED DESCRIPTION

[0061] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1-3 The present invention provides a technical solution: a transmission line fault early warning system driven by big data, the system comprising:

[0063] State prediction module: responsible for building and running the transmission line state prediction model. The model takes the historical operation data, environmental conditions and line parameters of the transmission line as input, and deeply mines and learns the input data through data analysis and processing technology, so as to output the future state indicator data of the transmission line in the future. In the model training stage, a large amount of historical operation data is first collected, including electrical parameters such as current, voltage, power factor, and historical environmental condition data, such as temperature, humidity, wind speed, etc., and combined with line parameters such as line length, conductor type, insulator model, etc., and the corresponding actual operation state data as training samples. Using these data, a model that can accurately predict the future state of the transmission line is trained through supervised learning or unsupervised learning methods.

[0064] Risk assessment module: responsible for building and running the fault risk assessment model. The model receives the real-time operation data, historical maintenance records, line parameters and environmental condition data of the transmission line as input, and evaluates the current fault risk index of the transmission line by comprehensively analyzing these data. In the model training stage, historical operation data, maintenance records (including the number of maintenance, maintenance type, maintenance time, etc.), line parameters, environmental condition data and corresponding fault records are collected. Using these data, through data mining techniques such as association rule mining and classification algorithms, a model that can accurately assess the risk of transmission line failure is trained. The model can comprehensively consider multiple factors, such as the degree of equipment aging, the severity of the environment, the frequency of historical failures, etc., so as to more comprehensively reflect the fault risk status of the transmission line.

[0065] Early warning decision module: responsible for building and running the fault early warning decision model. This model models the management process of the transmission line as a Markov decision process, and makes fault early warning decisions based on the output of the state prediction module (future state indicator data) and the output of the risk assessment module (current fault risk index).

[0066] Management execution module: responsible for real-time management based on the trained state prediction module, risk assessment module and early warning decision module. This module receives the output of the early warning decision module, that is, the optimal early warning strategy, and automatically executes early warning measures according to the strategy or transmits it to the management personnel for confirmation and execution through the management system.

[0067] In specific implementation, the management execution module can be integrated with the monitoring system, inspection system, communication system, etc. of the power system to realize functions such as automatic sending of warning information and automatic arrangement of inspection tasks. At the same time, the module can also provide a user interface to facilitate management personnel to view warning information and confirm warning measures.

[0068] The present invention will be further described below in conjunction with Examples 1 to 3:

[0069] Embodiment 1: Fault warning decision model The management process of the transmission line is modeled as a Markov decision process, which is specifically defined as follows:

[0070] State space: It is defined as the combination of real-time operation data, line parameters and environmental condition data of the transmission line. Real-time operation data includes but is not limited to electrical parameters such as current, voltage, power factor, etc.; line parameters include line length, conductor type, insulator model, etc.; environmental condition data includes temperature, humidity, wind speed, etc. These data together constitute the current state of the transmission line.

[0071] Action space: defined as a set of executable inspection, maintenance, adjustment and protection operations. For example, inspection operations may include regular inspections, special inspections, etc.; maintenance operations may include replacing damaged parts, repairing lines, etc.; adjustment operations may include adjusting line parameters, changing operating modes, etc.; protection operations may include switching on or off protection devices, etc.

[0072] Reward function: Calculates rewards or penalties based on the state prediction deviation and risk index deviation. Positive rewards are given to operations that reduce the deviation, and penalties are given to operations that increase the deviation. The specific implementation method of the reward function will be described in detail below.

[0073] Output strategy: It is defined as the method of selecting the best action in the current state. Through deep reinforcement learning algorithms, such as deep Q network (DQN), policy gradient algorithm, etc., the generated decision actions are continuously optimized, so that in the long-term operation process, the system can select the best warning and response measures.

[0074] The implementation method of the reward function includes the following steps:

[0075] (1) Obtaining real-time operating status data of the transmission line: Using sensors and monitoring systems, the operating data of the transmission line, including electrical parameters and environmental conditions, is collected in real time.

[0076] (2) Calculate the state prediction deviation: Compare the real-time operation state data with the predicted state index data output by the state prediction module and calculate the deviation between the two. For example, the difference between the measured value and the predicted value of electrical parameters such as current and voltage can be calculated, or the difference between the measured value and the predicted value of ambient temperature can be calculated.

[0077] (3) Obtaining real-time fault risk assessment data of the transmission line: Through the risk assessment module, the current fault risk index of the transmission line is calculated in real time.

[0078] (4) Calculate the risk index deviation: Compare the current fault risk index output by the fault risk assessment model with the preset risk standard or historical risk data to calculate the deviation of the risk index. For example, the difference between the current risk index and the average risk index can be calculated, or the difference between the current risk index and a specific risk threshold can be calculated.

[0079] (5) Calculate rewards or penalties based on state prediction deviation and risk index deviation: The specific formula is: Reward = -(α * state prediction deviation + β * risk index deviation), where α and β are the weight coefficients of the state prediction deviation and risk index deviation reward items, respectively, and α, β > 0. This means that when the state prediction deviation and risk index deviation decrease, the reward increases; when the deviation increases, the reward decreases (that is, the penalty increases). By adjusting the values ​​of α and β, the sensitivity of the system to the state prediction deviation and risk index deviation can be controlled.

[0080] Assume that at a certain moment, the real-time current of the transmission line is 100A, the predicted current is 95A, the measured risk index is 0.8, and the preset risk standard is 0.5. Then the state prediction deviation is |100-95|=5A, and the risk index deviation is |0.8-0.5|=0.3. If α=1 and β=2 are set, the reward is: - (15 + 20.3) = -5.6. This means that there is a deviation between the current state and the predicted state and the risk standard, and the system needs to take corresponding measures to reduce these deviations in order to obtain a higher reward.

[0081] Embodiment 2: The transmission line state prediction model is constructed using the convolutional long short-term memory network (ConvLSTM) algorithm. The structure, training steps and specific application of the ConvLSTM network in transmission line state prediction are described in detail below.

[0082] The ConvLSTM network is a deep learning model that combines a convolutional neural network (CNN) and a long short-term memory network (LSTM), and is particularly suitable for processing sequence data with spatiotemporal correlation. Its network structure includes:

[0083] Input layer: used to receive the training data of the model. In the prediction of transmission line status, the data received by the input layer include historical operation data (such as time series of electrical parameters such as current, voltage, power factor, etc.), environmental condition data (such as time series of temperature, humidity, wind speed, etc.), and line parameters (such as static information such as line length and conductor type). After preprocessing, these data are input into the ConvLSTM network in an appropriate form.

[0084] Hidden layer: contains at least one ConvLSTM hidden layer. Each ConvLSTM hidden layer consists of several ConvLSTM units, each of which has a forget gate, input gate, and output gate. These gating mechanisms are used to control the flow of information and state updates, allowing the ConvLSTM network to capture long-term dependencies in time series. At the same time, because the ConvLSTM unit is combined with convolution operations, it can consider spatial features and is suitable for processing data with spatial correlation.

[0085] Output layer: Receives the output of the hidden layer and maps the output of the ConvLSTM network to the predicted operating status indicator value through a fully connected layer. In the transmission line status prediction, the output layer may output the predicted values ​​of electrical parameters such as current, voltage, power factor, etc. in the future.

[0086] The steps of training the transmission line state prediction model include:

[0087] F101: Initialize the weights and biases of the ConvLSTM network

[0088] The weights and bias items of the ConvLSTM network are initialized using random initialization. Random initialization is a common method for deep learning model training. It helps break the symmetry and enables the model to learn different features during training.

[0089] F102: Select loss function

[0090] The mean square error (MSE) is selected as the loss function. MSE is a common indicator to measure the difference between the model prediction value and the true value. It calculates the average of the square of the difference between the predicted value and the true value. In the prediction of transmission line status, MSE can reflect the accuracy of the model in predicting electrical parameters (such as current, voltage, etc.).

[0091] F103: Use back propagation algorithm and Adam optimization algorithm to train the network

[0092] The gradient is calculated according to the loss function, and the back-propagation algorithm is used to pass the gradient to each layer of the network. Then, the Adam optimization algorithm is used to update the weights and bias terms of the ConvLSTM network. The Adam optimization algorithm is an adaptive learning rate optimization algorithm based on first-order and second-order moment estimation. It combines the advantages of the momentum method and the RMSProp method, and has the characteristics of fast convergence speed and strong robustness. The network is iteratively trained until the loss function converges, that is, the preset training goal is achieved.

[0093] F104: Save the trained ConvLSTM network model

[0094] After the training is completed, save the trained ConvLSTM network model. Saving the model can facilitate subsequent use and deployment, so that the model can be quickly loaded and state prediction can be performed in actual applications.

[0095] Take the prediction of the current of the transmission line in the next 24 hours as an example. First, collect historical operation data (including the current time series of the past period of time), environmental condition data (such as temperature, humidity, etc.) and line parameters. Then, preprocess these data and input them into the ConvLSTM network. By training the ConvLSTM network, the network can learn the law of current change over time and the relationship between environmental conditions and line parameters. Finally, use the trained ConvLSTM network model to predict the current in the next 24 hours and output the prediction results. :

[0096] Example 3: The fault risk assessment model uses a random forest algorithm, which improves the accuracy and robustness of the model by constructing multiple decision trees and integrating their results. The random forest algorithm can process high-dimensional data and has good tolerance for missing values ​​and outliers, and is suitable for fault risk assessment of transmission lines.

[0097] The steps to train a failure risk assessment model include:

[0098] F201: Constructing a training sample set for the random forest algorithm

[0099] Collect training data, including historical operation data (such as time series of electrical parameters such as current, voltage, power factor, etc.), maintenance records (such as inspection records, repair records, etc.), line parameters (such as line length, conductor type, insulator model, etc.) and environmental condition data (such as temperature, humidity, wind speed, etc.). Use these data as input features and the corresponding fault records (such as fault type, fault time, etc.) as labels to build a training sample set for the random forest algorithm.

[0100] In the specific implementation, the data needs to be preprocessed, including data cleaning (removing duplicate values, processing missing values, etc.), data conversion (such as converting categorical data into numerical data) and data normalization (converting data of different dimensions to the same dimension) to ensure the quality and consistency of the data.

[0101] F202: Use the random forest algorithm to train the training sample set

[0102] The constructed training sample set is trained using the random forest algorithm. The random forest algorithm constructs multiple decision trees by randomly selecting some input features and data samples. Each decision tree is a classifier or regressor that classifies or regresses the data based on the input features. During the training process, each decision tree will independently learn the characteristics and rules of the data and output its own prediction results.

[0103] The random forest algorithm combines the prediction results of all decision trees and obtains the final prediction result through voting. This integrated learning method can improve the accuracy and robustness of the model and reduce the risk of overfitting.

[0104] F203: Optimizing parameters of the random forest algorithm

[0105] During the training process, the parameters of the random forest algorithm, including the number of trees and the maximum depth, need to be optimized through cross-validation. Cross-validation is a method for evaluating model performance that divides the training data into multiple subsets, uses one of the subsets as the validation set each time, and uses the remaining subsets as the training set for training, and evaluates the performance of the model on the validation set.

[0106] Through multiple cross-validations, the performance of the model under different parameter combinations can be evaluated, and the optimal parameter combination can be selected as the parameters of the final model. Specifically, the number of trees can be increased and the changes in model performance can be observed until the performance is no longer significantly improved. At the same time, the maximum depth of the tree can also be adjusted to control the complexity of the model and avoid overfitting.

[0107] Take the evaluation of the risk of a transmission line failure in the future as an example. First, collect the historical operation data, maintenance records, line parameters and environmental condition data of the line, and build a training sample set. Then, use the random forest algorithm to train the training sample set and build a fault risk assessment model. During the training process, the parameters of the random forest algorithm, such as the number of trees and the maximum depth, are optimized by cross-validation. Finally, use the trained fault risk assessment model to evaluate the fault risk of the line in the future and output the evaluation results.

[0108] Example 4: This example is used to describe in detail the use of the PPO algorithm to build a transmission line fault warning decision model. The fault warning decision model uses the PPO algorithm, which is a reinforcement learning algorithm that can learn and optimize strategies in complex environments. The model includes two main networks: a policy network and a value network. The policy network is used to generate action probability distributions and decide what actions to take in a given state; the value network is used to estimate the state value, that is, to evaluate the quality of the current state.

[0109] The steps for training the fault warning decision model include:

[0110] S1: Initialize the parameters of the policy network and the value network. The policy network can be a neural network whose output layer represents the probability distribution of different actions. The value network is also a neural network whose output is a scalar that represents the valuation of the current state.

[0111] S2: Initialize the experience playback buffer D, set the buffer size N, and initialize the experience playback buffer D. The buffer is used to store the state transition experience generated during the power transmission line management process, including state s, action a, reward r, next state s' and whether the termination flag is done.

[0112] S3: For each training round

[0113] S301: Initialize the state s at the beginning of the round. At the beginning of each training round, the initialization state s is the current operating state of the transmission line or the initial state of the simulation.

[0114] S302: Before the round ends, the following steps are executed in a loop

[0115] i) Policy network selects action: The policy network selects action a based on the current state s and adds exploration noise to increase exploration. The exploration noise can be random noise or noise based on a certain distribution.

[0116] ii) Perform actions and observe results: Perform action a and observe the reward r, the next state s', and the termination flag done. The reward r is determined based on the effect of the action and the operation target of the transmission line.

[0117] iii) Storing experience: Store the experience (s, a, r, s', done) in the experience replay buffer D for subsequent use in updating network parameters.

[0118] iv) Randomly sample experience: Randomly sample a batch of experience from the experience replay buffer D to update the network parameters. Random sampling helps break the temporal correlation of experience and improve the stability of training.

[0119] v) Update the value network parameters: Use the sampled experience to update the parameters of the value network to minimize the prediction error of the value function. Gradient descent or other optimization algorithms can be used to update the parameters.

[0120] vi) Update policy network parameters: Use the sampled experience to update the policy network parameters. The objective function of the PPO algorithm is optimized, which takes into account the update amplitude and stability of the policy to ensure a smooth transition of the policy.

[0121] vii) Update the current state: Update the current state s to the next state s' to continue the next round of action selection and execution.

[0122] S303: When the round ends (ie done is True), reset the state s to a new round start state to start the next round of training.

[0123] S4: During the training process, the performance of the fault warning decision model is regularly evaluated. The cumulative reward value is used as the performance evaluation indicator for the current round of training.

[0124] S401: After each training round, the cumulative reward value of the round is calculated as the performance evaluation indicator of the current round iteration. The cumulative reward value reflects the overall performance of the model in the round.

[0125] S402: If the cumulative reward value of the current round of iteration is higher than the cumulative reward value of the previous round of iteration, it indicates that the model performance has improved. The current model parameters are kept unchanged and the next round of iteration is continued.

[0126] S403: If the cumulative reward value of the current iteration is lower than the cumulative reward value of the previous iteration, it indicates that the model performance has declined, and it is necessary to adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise, and the size of the experience playback buffer. After adjusting the parameters, reinitialize the parameters of the policy network and the value network, and start a new round of training.

[0127] S5: Repeat steps S3 and S4 until the model performance reaches a preset standard. The preset standard can be determined according to the actual application scenario and requirements, such as the threshold of the cumulative reward value, the accuracy of the model on the test set, etc.

[0128] Taking the fault warning of transmission lines as an example, the state s can represent the current operating parameters of the transmission line (such as current, voltage, temperature, etc.), the action a can represent the maintenance measures taken (such as inspection, repair, etc.), and the reward r can be determined based on factors such as whether the fault occurs and the maintenance cost. By training the fault warning decision model, the model can learn what actions to take in different states to maximize the cumulative reward value, thereby achieving fault warning and decision support.

[0129] The present invention also includes a transmission line fault early warning method based on big data drive, the method comprising:

[0130] F1: Construct a transmission line state prediction model to predict the operation state of the transmission line in the future. Its input data includes: historical operation data, environmental conditions and line parameters of the transmission line. Its output data is: future state indicator data of the transmission line. Model training data includes: historical operation data, historical environmental conditions data, line parameters and corresponding actual operation state data.

[0131] F2: Construct a fault risk assessment model to assess the current fault risk of the transmission line; its input data includes: real-time operation data of the transmission line, historical maintenance records, line parameters and environmental condition data; its output data is: the current fault risk index of the transmission line; model training data includes: historical operation data, maintenance records, line parameters, environmental condition data and corresponding fault records;

[0132] F3: Construct a fault warning decision model, model the management process of the transmission line as a Markov decision process, and make fault warning decisions based on the output of the transmission line state prediction model and the output of the fault risk assessment model;

[0133] F4: Real-time management is carried out based on the trained transmission line status prediction model, fault risk assessment model and fault warning decision model, and the decision results are automatically executed or transmitted to the management personnel for confirmation and execution through the management system.

[0134] The implementation of this method refers to the above embodiment and will not be described in detail in the specification.

[0135] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0136] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A transmission line fault early warning system driven by big data, characterized in that: The system comprises: The state prediction module is used to build and run a transmission line state prediction model, which receives historical operation data, environmental conditions and line parameters of the transmission line as input, and outputs future state indicator data of the transmission line in the future period of time; when training the model, the historical operation data, historical environmental condition data, line parameters and corresponding actual operation state data are used as training data; The risk assessment module is used to build and run a fault risk assessment model, which receives the real-time operation data, historical maintenance records, line parameters and environmental condition data of the transmission line as input, and outputs the current fault risk index of the transmission line; when training the model, the historical operation data, maintenance records, line parameters, environmental condition data and corresponding fault records are used as training data; The early warning decision module is used to build and run the fault early warning decision model, which models the management process of the transmission line as a Markov decision process and makes fault early warning decisions based on the output of the state prediction module and the output of the risk assessment module; The management execution module is used to perform real-time management based on the trained status prediction module, risk assessment module and early warning decision module, and automatically execute the decision results or transmit them to the management personnel for confirmation and execution through the management system; The transmission line state prediction model is constructed using a convolutional long short-term memory network ConvLSTM algorithm; the steps of training the transmission line state prediction model include: F101: Initialize the weights and biases of the ConvLSTM network using a random initialization method; F102: Select the loss function to measure the difference between the model prediction value and the true value. The mean square error MSE is used as the loss function. F103: Use the back propagation algorithm and Adam optimization algorithm to calculate the gradient according to the loss function, update the weights and bias items of the ConvLSTM network, and iteratively train the network until the loss function converges; F104: Save the trained ConvLSTM network model.

2. The power transmission line fault early warning system based on big data drive according to claim 1 is characterized in that: The specific definition of the fault warning decision model includes: State space: defined as the combination of real-time operation data of the transmission line, line parameters and environmental condition data; Action space: defined as the set of executable inspection, maintenance, adjustment and protection operations; Reward function: Calculate rewards or penalties based on state prediction deviation and risk index deviation, give positive rewards to operations that reduce deviations, and give penalties to operations that increase deviations; Output strategy: It is defined as the method of selecting the best action in the current state, and the decision actions generated are continuously optimized through deep reinforcement learning algorithms.

3. The power transmission line fault early warning system based on big data drive according to claim 2 is characterized in that: The implementation method of the reward function includes: Obtain real-time operation status data of the transmission line; compare the real-time operation status data with the predicted status indicator data, and calculate the status prediction deviation; Obtain real-time fault risk assessment data of the transmission line; compare the current fault risk index output by the fault risk assessment model with the preset risk standard or historical risk data, and calculate the risk index deviation; Rewards or penalties are calculated based on the state prediction deviation and risk index deviation. The specific formula is: Reward = - (α * state prediction deviation + β * risk index deviation), where α and β are the weight coefficients of the state prediction deviation and risk index deviation reward items respectively, and α, β > 0.

4. The power transmission line fault early warning system based on big data drive according to claim 1 is characterized in that: The real-time operation data includes current, voltage, power factor, temperature and line impedance; the environmental condition data includes ambient temperature, humidity, wind speed, wind direction and lightning activity; the line parameters include line length, material, cross-sectional area and installation height; the maintenance records include maintenance time, maintenance type, maintenance cost and maintenance effect; The executable inspection, maintenance, adjustment and protection operation set includes: line inspection, component replacement, tightening connections, adjusting line tension, installing lightning protection equipment and adjusting protection device settings.

5. The power transmission line fault early warning system based on big data drive according to claim 1 is characterized in that: The network structure of the transmission line state prediction model includes: Input layer: used to receive training data of the model; Hidden layer: contains at least one ConvLSTM hidden layer. Each ConvLSTM hidden layer contains several ConvLSTM units. Each ConvLSTM unit has a forget gate, input gate and output gate to control the flow of information and state update, while considering spatial features. Output layer: Receives the output of the hidden layer and maps the output of the ConvLSTM network to the predicted operating status indicator value through a fully connected layer.

6. The power transmission line fault early warning system based on big data drive according to claim 1 is characterized in that: The fault risk assessment model is constructed using the random forest RF algorithm, and its training steps include: F201: Use the historical operation data, maintenance records, line parameters and environmental condition data in the training data as input features and the corresponding fault records as labels to construct a training sample set for the random forest algorithm; F202: Use the random forest algorithm to train the training sample set, and improve the accuracy and robustness of the model by building multiple decision trees and combining their results; F203: During the training process, the number and maximum depth of trees of the random forest algorithm are optimized by cross-validation.

7. The power transmission line fault early warning system based on big data drive according to claim 1 is characterized in that: The fault warning decision model is constructed using the proximal strategy optimization PPO algorithm, and its training steps include: S1. Initialize the policy network and value network: the policy network is used to generate action probability distribution, and the value network is used to estimate state value; S2. Initialize the experience playback buffer D: set the buffer size N to store the state transition experience generated during the power transmission line management process, including state s, action a, reward r, next state s' and whether the termination flag is done; S3. For each training round: S301: Initialize the state s at the beginning of the round; S302: Before the round is terminated, the following steps are executed in a loop: i) the policy network selects action a according to the current state s and adds exploration noise; ii) executes action a, observes the reward r, the next state s' and the termination flag done; iii) stores the experience (s, a, r, s', done) in the experience replay buffer D; iv) randomly samples a batch of experience from the experience replay buffer D to update the network parameters; v) uses the sampled experience to update the parameters of the value network to minimize the prediction error of the value function; vi) uses the sampled experience to update the parameters of the policy network and optimizes it through the objective function of the PPO algorithm; vii) updates the current state s to the next state s'; S303: When the round ends, reset the state s to the new round start state; S4: During the training process, the performance of the fault warning decision model is regularly evaluated, and based on the evaluation results, the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise, and the size of the experience replay buffer are adjusted; S5: Repeat steps S3 and S4 until the model performance reaches a preset standard.

8. The power transmission line fault early warning system based on big data drive according to claim 7 is characterized in that: In step S4, the accumulated reward value is used as the performance evaluation indicator of the current round of training, and the specific implementation steps include: S401: After each training round, the cumulative reward value of the round is calculated as the performance evaluation indicator of the current round iteration; S402: If the cumulative reward value of the current iteration is higher than the cumulative reward value of the previous iteration, the current model parameters are kept unchanged and the next iteration is continued; S403: If the cumulative reward value of the current round of iterations is lower than the cumulative reward value of the previous round of iterations, adjust the weight coefficient of the reward function, the learning rate of the model, the size of the exploration noise and the size of the experience replay buffer, reinitialize the parameters of the policy network and the value network, and start a new round of training.

9. A transmission line fault early warning method based on big data drive, characterized in that: The method comprises: F1: Construct a transmission line state prediction model to predict the operation state of the transmission line in the future. Its input data includes: historical operation data, environmental conditions and line parameters of the transmission line. Its output data is: future state indicator data of the transmission line. Model training data includes: historical operation data, historical environmental conditions data, line parameters and corresponding actual operation state data. F2: Construct a fault risk assessment model to assess the current fault risk of the transmission line; its input data includes: real-time operation data of the transmission line, historical maintenance records, line parameters and environmental condition data; its output data is: the current fault risk index of the transmission line; model training data includes: historical operation data, maintenance records, line parameters, environmental condition data and corresponding fault records; F3: Construct a fault warning decision model, model the management process of the transmission line as a Markov decision process, and make fault warning decisions based on the output of the transmission line state prediction model and the output of the fault risk assessment model; F4: Real-time management is performed based on the trained transmission line status prediction model, fault risk assessment model and fault warning decision model, and the decision results are automatically executed or transmitted to the management personnel for confirmation and execution through the management system; The transmission line state prediction model is constructed using a convolutional long short-term memory network ConvLSTM algorithm; the steps of training the transmission line state prediction model include: F101: Initialize the weights and biases of the ConvLSTM network using a random initialization method; F102: Select the loss function to measure the difference between the model prediction value and the true value. The mean square error MSE is used as the loss function. F103: Use the back propagation algorithm and Adam optimization algorithm to calculate the gradient according to the loss function, update the weights and bias items of the ConvLSTM network, and iteratively train the network until the loss function converges; F104: Save the trained ConvLSTM network model.

Images