Intelligent line fault prediction and identification system based on AI algorithm

Abstract

The invention provides an intelligent line fault prediction and identification system based on an AI algorithm, and relates to the technical field of power transmission line fault prediction. The system comprises a data acquisition terminal, a data analysis platform and a data processing terminal which are in communication connection with one another, the data acquisition terminal is used for acquiring to-be-analyzed data of the power transmission line according to a preset acquisition time requirement and sending the to-be-analyzed data to the data analysis platform; the data analysis platform is used for receiving the to-be-analyzed data, inputting the to-be-analyzed data into the line fault prediction model to obtain a fault prediction result of the power transmission line, and sending the fault prediction result to the data processing terminal; wherein the line fault prediction model is a model obtained by training a deep convolutional neural network and a support vector machine; and the data processing terminal is used for receiving the fault prediction result and generating a processing strategy and early warning information for the power transmission line based on the fault prediction result. The system can improve the fault prediction effect of the power transmission line, and reduces the loss caused by the fault.

Description

Technical Field

[0001] The present application relates to the technical field of power transmission line fault prediction, and in particular to an intelligent line fault prediction and identification system based on AI algorithm. Background Art

[0002] As the demand for electricity grows, the safe and stable operation of transmission lines, as an important channel for power transmission, is crucial to the entire power system. Line failures account for 90% of electrical failures, and traditional fault diagnosis and early warning methods rely on manual experience and expertise, which can no longer meet the requirements of high efficiency and quality.

[0003] Existing power line fault diagnosis systems usually include two main subsystems: a fault information collection subsystem and a fault type diagnosis subsystem. The fault information collection subsystem is responsible for collecting relevant data from power equipment and classifying it, while the fault type diagnosis subsystem uses artificial intelligence (AI) and other technologies to determine the specific fault type that occurred in the power system based on the input equipment data and provide a basis for solving the problem. In addition, deep learning technology, especially convolutional neural networks, plays an important role in transmission line fault prediction, and can automatically extract features from data to improve the accuracy of fault prediction.

[0004] However, the existing methods still have the problem of poor prediction effect when predicting transmission line faults. Summary of the invention

[0005] The present application provides an intelligent line fault prediction and identification system based on AI algorithm to solve the problem that the existing methods have poor prediction effect when predicting transmission line faults.

[0006] In a first aspect, the present application provides an intelligent line fault prediction and identification system based on an AI algorithm, comprising:

[0007] Data collection terminals, data analysis platforms and data processing terminals that are interconnected and communicate with each other;

[0008] The data collection terminal is used to collect the data to be analyzed of the transmission line according to the preset collection time requirements and send it to the data analysis platform. The data to be analyzed includes the power data of the transmission line and the environmental data of the location where the transmission line is located;

[0009] The data analysis platform is used to receive the data to be analyzed, and input the data to be analyzed into the line fault prediction model to obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal; wherein the line fault prediction model is a model obtained by training a deep convolutional neural network and a support vector machine;

[0010] The data processing terminal is used to receive the fault prediction results, and based on the fault prediction results, generate processing strategies and early warning information for the transmission line.

[0011] In a possible implementation, the data analysis platform includes a data acquisition module and a model processing module that are communicatively connected to each other;

[0012] The data acquisition module is used to receive the data to be analyzed sent by the data acquisition terminal;

[0013] The model processing module is used to input the data to be analyzed into the line fault prediction model, obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal;

[0014] The model processing module is also used to train the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed, until the fireworks fitness value set meets the preset optimization conditions, and the line fault prediction model is obtained;

[0015] Among them, the initial line fault prediction model is constructed based on deep convolutional neural network and support vector machine, and the fireworks fitness value set is determined by optimizing the parameters in the initial line fault prediction model according to the preset fireworks algorithm.

[0016] In a possible implementation, the model processing module is further used to:

[0017] Before training the pre-built initial line fault prediction model, building an initial line fault prediction model, the initial line fault prediction model includes a first initial prediction model consisting of an input layer, a convolution layer, a pooling layer, and a fully connected layer, and a second initial prediction model consisting of a support vector machine classification layer;

[0018] Among them, in the first initial prediction model, the input layer is used to receive the data to be analyzed or the historical data to be analyzed, the convolution layer is used to extract the local features of the data in the input layer, the pooling layer is used to transform the dimensions of the local features to obtain the transformed local features, and the fully connected layer is used to obtain the global features according to the transformed local features;

[0019] In the second initial prediction model, the support vector machine classification layer is connected behind the fully connected layer in the first initial prediction model, and is used to receive the global features output by the first prediction model and classify the global features to predict whether a transmission line fault occurs.

[0020] In a possible implementation, the model processing module trains the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed until the fireworks fitness value set meets the preset optimization conditions, and obtains the line fault prediction model, which is specifically used to:

[0021] According to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection result corresponding to each data to be analyzed, a sample set is constructed, wherein each sample in the sample set includes: the data to be analyzed collected during a fault prediction, and the actual fault detection result corresponding to this fault prediction;

[0022] Based on the sample set and the preset gradient optimization algorithm, a first initial prediction model in the initial line fault prediction model is trained to obtain a first prediction model, where the first prediction model is a model obtained by optimizing model parameters of the first initial prediction model based on the preset gradient optimization algorithm;

[0023] Determine, according to the first prediction model, a global feature corresponding to each sample in the sample set;

[0024] Based on the actual fault detection result of each sample in the sample set and the corresponding global feature, the second initial prediction model in the initial line fault prediction model is trained until the fireworks fitness value set meets the preset optimization condition, thereby obtaining the second prediction model;

[0025] A line fault prediction model is obtained according to the first prediction model and the second prediction model.

[0026] In a possible implementation, the model processing module trains the second initial prediction model in the initial line fault prediction model based on the actual fault detection result of each sample in the sample set and the corresponding global feature until the fireworks fitness value set meets the preset optimization condition and the second prediction model is obtained, specifically for:

[0027] Taking the parameters to be optimized of the second initial prediction model as optimization samples, and determining the number and value range of the optimization samples;

[0028] Determine the values ​​of multiple optimized samples according to the number and value range of the optimized samples;

[0029] For each optimized sample, the global feature corresponding to each sample in the sample set is input into the second initial prediction model corresponding to the optimized sample to obtain the fault prediction result corresponding to each sample in the sample set;

[0030] Determine a first fireworks fitness value of the optimized sample according to the actual fault detection result and the corresponding fault prediction result of each sample in the sample set, wherein the fireworks fitness value is used to indicate the accuracy of the second initial prediction model corresponding to the optimized sample in predicting the transmission line fault;

[0031] According to the first firework fitness value of each optimized sample, all optimized samples are subjected to explosion processing to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample;

[0032] For each optimized sample, each spark in the optimized sample is used as a new optimized sample, and the second firework fitness value of each new optimized sample is determined;

[0033] Determine a plurality of target optimized samples and a fireworks fitness value of each target optimized sample according to the first fireworks fitness value of each optimized sample and the second fireworks fitness value of each new optimized sample;

[0034] If the fireworks fitness values ​​of all target optimized samples meet the preset accuracy requirement, a final optimized sample is determined from all target optimized samples, wherein the model corresponding to the final optimized sample is the second prediction model.

[0035] In a possible implementation, when the model processing module performs explosion processing on all optimized samples according to the first fireworks fitness value of each optimized sample to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample, it is specifically used to:

[0036] According to the first firework fitness value of each optimized sample, all optimized samples are subjected to explosion processing to obtain the initial firework explosion radius and number of sparks corresponding to each optimized sample;

[0037] Based on the Tent chaotic map, the range and distribution characteristics of the initial fireworks explosion radius are adjusted to obtain the fireworks explosion radius.

[0038] In a possible implementation, the data analysis platform further includes a data preprocessing module, which is respectively communicated with the data acquisition module and the model processing module;

[0039] The data preprocessing module is used to decompose the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, and the decomposed data to be analyzed is the characteristic data in the frequency domain;

[0040] The data preprocessing module is also used to normalize the decomposed data to be analyzed to obtain preprocessed data to be analyzed, wherein the data to be analyzed in the model processing module is the preprocessed data to be analyzed.

[0041] In a possible implementation manner, when the data preprocessing module decomposes the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, it is specifically used to:

[0042] Perform three-layer decomposition processing on the data to be analyzed in the data acquisition module to determine the energy of the data to be analyzed in each frequency band and the total energy in all frequency bands;

[0043] According to the energy of each frequency band and the total energy in all frequency bands, determine the energy proportion of each frequency band and the energy entropy corresponding to the energy proportion;

[0044] According to the energy entropy of each frequency band, the decomposed data to be analyzed is determined.

[0045] In a possible implementation, the data analysis platform further includes: a model evaluation module;

[0046] The model evaluation module is used to evaluate the prediction performance of the line fault prediction model based on preset evaluation indicators after the model processing module obtains the line fault prediction model. The preset evaluation indicators include the accuracy, precision, recall rate, and the harmonic mean of the precision and recall rate of the model prediction.

[0047] In a possible implementation, the system further includes: a control module;

[0048] The control module is used to perform fault processing operations on the transmission line based on the processing strategy generated by the data processing terminal;

[0049] The control module is also used to receive processing instructions determined by maintenance personnel based on the early warning information, and perform fault processing operations on the transmission line according to the processing instructions.

[0050] The present application provides an intelligent line fault prediction and identification system based on AI algorithm. The data acquisition terminal collects the data to be analyzed of the transmission line according to the preset collection time requirement and sends it to the data analysis platform. The data to be analyzed includes the power data of the transmission line and the environmental data of the location of the transmission line. After receiving the data to be analyzed, the data analysis platform inputs the data to be analyzed into the line fault prediction model to obtain the fault prediction result of the transmission line, and sends the fault prediction result to the data processing terminal. Among them, the line fault prediction model is a model obtained by training the deep convolutional neural network and the support vector machine. After receiving the fault prediction result, the data processing terminal will generate a processing strategy for the transmission line based on the fault prediction result and issue a warning information, so as to use the deep convolutional neural network to extract the data features of various data of the transmission line, and then use the support vector machine for prediction. The combination of the two can more accurately predict the fault situation of the transmission line, and then reasonably generate the processing strategy and warning information according to the accurate prediction result, so as to prevent possible faults of the transmission line in advance, reduce the losses caused by the fault, and improve the safety and stability of the operation of the transmission line. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0052] Figure 1 A schematic diagram of the structure of an intelligent line fault prediction and identification system based on AI algorithm provided in this application;

[0053] Figure 2 A schematic diagram of the modules of the data analysis platform provided for this application;

[0054] Figure 3 A schematic diagram of the operation flow of an intelligent line fault prediction and identification system based on AI algorithm provided in this application;

[0055] Figure 4 A schematic diagram of the structure of the electronic device provided in this application.

[0056] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0057] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0058] First, the terms involved in this application are explained:

[0059] Convolutional Neural Network (CNN): A deep learning model designed specifically for processing grid-structured data (such as images, audio). It automatically extracts features from data through components such as convolutional layers, pooling layers, and fully-connected layers. DC-CNN can be a deep convolutional neural network, which usually has more convolutional layers and pooling layers than CNN and can learn more complex and abstract features.

[0060] Support Vector Machine (SVM): A supervised learning classification algorithm whose basic idea is to find an optimal hyperplane in the feature space to separate samples of different categories as much as possible. For linearly separable data, this hyperplane can maximize the interval between two types of data. The interval refers to the distance from the hyperplane to the nearest sample point (support vector).

[0061] Fireworks Algorithm (FWA): A heuristic optimization algorithm inspired by the process of fireworks exploding in the night sky. In this algorithm, each firework represents a solution, and the explosion of fireworks is equivalent to searching in the solution space to obtain the optimal solution.

[0062] In modern society, the continuous rise in electricity demand has led to an increasing number of transmission lines. As the core channels for power transmission, these transmission lines not only bear huge electrical energy loads, but are also the key to maintaining the safe and stable operation of the entire power system. Therefore, the fact that line failures often rank first among electrical failures cannot be ignored.

[0063] In the prior art, line faults are generally judged and analyzed by relying on human experience and professional knowledge. Usually, the power line fault diagnosis system includes two main subsystems, namely the fault information collection subsystem and the fault type diagnosis subsystem. The fault information collection subsystem is responsible for collecting relevant data of power equipment and classifying it, while the fault type diagnosis subsystem needs to design corresponding fault identification rules or models based on the long-term accumulated experience and professional knowledge reserves of humans, determine the specific fault type occurring in the power system based on the input equipment data, and provide a basis for solving the problem. However, this method has the following problems:

[0064] 1. Data collection is limited by factors such as equipment type, environmental conditions, and sensor quality, resulting in insufficient or low-quality available data. The accuracy of fault prediction is greatly affected by data collection.

[0065] 2. It is easy to overfit when building a model, that is, the model is too complex to perform well on the training set, but the prediction accuracy on new samples is not high and the model generalization ability is weak.

[0066] 3. For fault prediction and diagnosis tasks of large-scale equipment, the computational efficiency is low and cannot meet the real-time requirements.

[0067] 4. Effective fault identification rules or models can only be designed by relying on human experience and expertise, but the development and upgrading of equipment technology is very fast, and old experience may no longer be applicable.

[0068] In the face of the above technical problems, the present application provides an intelligent line fault prediction and identification system based on AI algorithm. By setting up a data acquisition terminal and clearly collecting data related to the transmission line according to the preset acquisition time requirements, the power data of the transmission line itself and the environmental data of the location are fully considered. This multi-dimensional data acquisition provides sufficient and comprehensive basic information for subsequent accurate fault prediction; a deep convolutional neural network is used to automatically extract complex features in the data, and potential, deep-level patterns and associations are mined from a large amount of power data and environmental data, and then a support vector machine is used for classification and prediction. The combination of the two can more accurately predict the fault conditions of the transmission line, and then based on the accurate prediction results, reasonable processing strategies and early warning information are generated, which can prevent possible faults in the transmission line in advance, reduce the losses caused by faults, and improve the safety and stability of transmission line operation.

[0069] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0070] Reference Figure 1 , Figure 1 A structural diagram of an intelligent line fault prediction and identification system based on AI algorithm provided for this application.

[0071] The above-mentioned intelligent line fault prediction and identification system based on AI algorithm includes a data acquisition terminal, a data analysis platform and a data processing terminal which are connected to each other in communication;

[0072] The data collection terminal is used to collect the data to be analyzed of the transmission line according to the preset collection time requirements and send it to the data analysis platform. The data to be analyzed includes the power data of the transmission line and the environmental data of the location where the transmission line is located;

[0073] The data analysis platform is used to receive the data to be analyzed, and input the data to be analyzed into the line fault prediction model to obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal; wherein the line fault prediction model is a model obtained by training a deep convolutional neural network and a support vector machine;

[0074] The data processing terminal is used to receive the fault prediction results, and based on the fault prediction results, generate processing strategies and early warning information for the transmission line.

[0075] The preset collection time requirement may refer to the time parameter set in the data collection terminal, which is used to determine when to collect data. These requirements can be set according to the needs of specific application scenarios to ensure the timeliness and effectiveness of data collection. For example, through the management software of the data collection terminal, the user can enter specific time parameters in the software interface; send configuration instructions to the data collection terminal through the data analysis platform to remotely update the collection time requirements; use machine learning or artificial intelligence algorithms to automatically adjust the collection time requirements based on historical data and prediction models to optimize the data collection strategy.

[0076] Furthermore, the data acquisition terminal collects data from the transmission line at preset time intervals (such as every 15 minutes). For example, in terms of power data, it will collect real-time voltage values, current values, power factors, etc. on the line. Assuming that a transmission line has a stable voltage of 220kV, a current of about 500A, and a power factor of 0.95 during normal operation. When the line is abnormal, these values ​​may fluctuate, such as the voltage suddenly drops to 180kV, the current increases to 600A, etc. For environmental data, the data acquisition terminal uses temperature sensors, humidity sensors, anemometers and other equipment to collect environmental information at the location of the line. For example, in mountainous transmission lines, in heavy rain weather, the humidity in the environmental data will rise sharply to more than 90%, and the wind speed may reach more than 20m / s. These environmental changes may affect the line and be reflected in the collected data. The collected power data and environmental data will be transmitted to the data analysis platform in a timely manner, such as by 4G\5G\APN and other methods.

[0077] After receiving the data, the data analysis platform inputs it into the line fault prediction model. This model is trained based on deep convolutional neural networks and support vector machines. The deep convolutional neural network first extracts features from the data. For example, it can extract potential feature patterns from the waveform changes of power data and the numerical changes of environmental data; and the support vector machine makes classification decisions based on these extracted features to determine whether there is a fault in the line and the type of fault. For example, if the feature data indicates that the line has a characteristic pattern of insulation aging and partial discharge, the support vector machine will classify it as an insulation fault type and output a fault prediction result, such as "insulation fault, partial discharge, failure probability 70%" and other information, and then send this result to the data processing terminal.

[0078] After receiving the fault prediction results, the data processing terminal generates processing strategies and warning information according to preset rules. Among them, the fault prediction results, processing strategies and warning information of the data processing terminal can be displayed to the staff through the local area network or mobile wireless network, such as a large visual screen, SMS alarm, APP warning, etc. For example, if the fault prediction result is "line overheating fault, fault probability 80%", the data processing terminal may generate such a processing strategy: first adjust the load distribution of the line and transfer part of the load to other parallel lines; then send maintenance personnel with infrared thermometers and other equipment to the faulty line section for inspection; and send warning information to the power dispatching center and relevant operation and maintenance personnel. The content of the warning information can be "Line [specific number] has an overheating fault risk, with a fault probability of 80%, and load adjustment measures have been taken. Please ask relevant personnel to check and handle it in time."

[0079] In some examples, the data processing terminal may be specifically used to:

[0080] 1) Receive the line fault prediction results transmitted by the data analysis platform. The prediction results may include the probability of fault occurrence, possible fault types (such as short circuit, open circuit, leakage, etc.) and the location information where the fault may occur (accurate to the line segment or key node).

[0081] 2) Based on the mapping table between the preset fault type and the processing strategy, a preliminary processing strategy is generated for the predicted fault type. For example, if a short circuit fault is predicted and located in a specific area of ​​a branch line, the preliminary processing strategy may be to first try to cut off the power supply of the branch line, start the backup line for power supply switching, and start the fault detection program to conduct detailed detection of the short circuit area to determine the cause of the short circuit.

[0082] 3) Optimize the preliminary processing strategy in combination with the fault probability information in the prediction results. If the fault probability is high (such as more than 80%), the preliminary processing strategy is directly determined as the final processing strategy and marked as a high-priority emergency processing strategy; if the fault probability is at a medium level (such as 30%-80%), some auxiliary detection and data collection steps are added to the preliminary processing strategy to further confirm the fault situation during the execution of the processing strategy, and it is marked as a medium-priority strategy; if the fault probability is low (less than 30%), an early warning message is generated and sent to the maintenance personnel for manual review and decision-making, and the processing strategy is marked as a low-priority observation strategy.

[0083] 4) The finalized processing strategy (high and medium priority) and warning information (low priority) are displayed to maintenance personnel or other users.

[0084] Optionally, the data acquisition terminal can be an IoT smart sensor or a drone equipped with a measuring sensor. These devices are installed at preset distances along the transmission line, such as every 15-25 kilometers, to collect and acquire the data to be analyzed, wherein the power data of the transmission line in the data to be analyzed may include current, voltage, frequency factor, etc.; the environmental data at the location of the transmission line may include wind speed, wind direction frequency, temperature, humidity, etc.

[0085] Optionally, the data to be analyzed may also include the physical state of the transmission line, such as the state of the transmission line being exposed, broken, and overheated. These abnormal physical states will increase the possibility of wind-induced discharge of the line, thereby exacerbating the line failure. The physical state of the transmission line can be collected by installing a high-definition camera, infrared thermal imager, and temperature and humidity sensor on the drone.

[0086] The intelligent line fault prediction and identification system based on AI algorithm provided in the embodiment of the present application improves the prediction accuracy through the combination of deep convolutional neural network and support vector machine, and can detect potential fault hazards such as line overload in advance, and has stronger generalization ability than traditional judgment methods; the data processing terminal promotes the transformation of intelligent operation and maintenance from fault prediction to proactive response, such as the arrangement of protective measures before severe weather, effectively reducing the probability of faults and losses and improving the level and efficiency of intelligent operation and maintenance, thereby enhancing system reliability and significantly reducing the failure rate of transmission lines.

[0087] On the basis of the above embodiment, the data analysis platform includes a data acquisition module and a model processing module which are communicatively connected to each other; the data acquisition module is used to receive the data to be analyzed sent by the data acquisition terminal; the model processing module is used to input the data to be analyzed into the line fault prediction model, obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal; the model processing module is also used to train the pre-constructed initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection result corresponding to each data to be analyzed, until the fireworks fitness value set meets the preset optimization conditions, and the line fault prediction model is obtained; wherein, the initial line fault prediction model is constructed based on a deep convolutional neural network and a support vector machine, and the fireworks fitness value set is determined by optimizing the parameters in the initial line fault prediction model according to a preset fireworks algorithm.

[0088] Among them, the function of the data acquisition module is to receive the data to be analyzed collected from the data acquisition terminal. This is the data source of the entire data analysis process and provides basic materials for subsequent analysis.

[0089] On the one hand, the model processing module inputs the received data to be analyzed into the line fault prediction model to obtain the fault prediction result of the transmission line, and transmits the result to the data processing terminal for relevant personnel to view and process; on the other hand, the initial line fault prediction model is trained using the data to be analyzed of the transmission line in the preset historical time period and the corresponding actual fault detection results. The training process optimizes the model parameters based on the preset fireworks algorithm, and measures the degree of optimization with the fireworks fitness value set until the final line fault prediction model is obtained after the preset optimization conditions are met. Among them, the initial line fault prediction model is constructed by combining the powerful feature extraction capability of the deep convolutional neural network and the good classification performance of the support vector machine. The fireworks algorithm is an intelligent algorithm for optimizing model parameters, and the fireworks fitness value set is an indicator for evaluating the effect of parameter optimization.

[0090] Optionally, the actual fault detection result may be a result obtained by a maintenance personnel performing a patrol inspection on the transmission line or a detection result based on an early warning issued by a monitor of the transmission line.

[0091] By constructing a line fault prediction model based on deep convolutional neural networks and support vector machines, and using the preset fireworks algorithm to optimize and train its parameters, the accuracy and reliability of transmission line fault prediction can be effectively improved. Compared with the existing technology, it has significant advantages in fault prediction accuracy, efficiency and model adaptability, and can better meet the power system's needs for transmission line fault prediction and ensure the stable operation of the power system.

[0092] On the basis of the above embodiment, the model processing module is also used to: before training the pre-constructed initial line fault prediction model, construct an initial line fault prediction model, the initial line fault prediction model includes a first initial prediction model consisting of an input layer, a convolution layer, a pooling layer, and a fully connected layer, and a second initial prediction model consisting of a support vector machine classification layer; wherein, in the first initial prediction model, the input layer is used to receive data to be analyzed or historical data to be analyzed, the convolution layer is used to extract local features of the data in the input layer, the pooling layer is used to perform dimension conversion on the local features to obtain converted local features, and the fully connected layer is used to obtain global features based on the converted local features; in the second initial prediction model, the support vector machine classification layer is connected after the fully connected layer in the first initial prediction model, and is used to receive the global features output by the first prediction model and classify the global features to predict whether a transmission line fault occurs.

[0093] In one example, the first initial prediction model is mainly composed of DC-CNN, including 1 input layer (2 channels, inputting power data and environmental data respectively), 2×2 convolutional layers, 2×2 pooling layers, 2 fully connected layers and 1 output layer (used for transition when connecting with the support vector machine classification layer (mainly composed of SVM), that is, the data of the output layer is the input data of the support vector machine classification layer). It should be noted that when the data to be analyzed also includes physical state, the channels and fully connected layers of the input layer can be 3, and the convolutional layer and pooling layer are 3×3, which are used to extract key features of different types of data to be analyzed.

[0094] Optionally, the first initial prediction model can also add an attention layer between the pooling layer and the fully connected layer, which is used to dynamically adjust the weight of the features based on the attention mechanism algorithm according to the corresponding geographical environment when the data to be analyzed is collected (such as the river area, mountainous area and other different geographical environments where the transmission line is located). Taking the transmission line in different geographical environments as an example, in the river area, due to the high water vapor content, the temperature and humidity may have a greater impact on the operation status of the line. The attention layer can give higher weights to temperature and humidity; in the mountainous area, the terrain is complex, and the impact of temperature changes and wind direction and speed on line safety may be more critical. The attention layer can highlight the weight proportion of these factors accordingly. Adding an attention layer can make the model more flexible to adapt to different operating environments. The environment in which the transmission line is located is complex and diverse, and the importance of fault factors in different environments is different. After adding the attention layer, the model can better focus on the environmental data that has a greater impact on faults in a specific environment, thereby improving the accuracy of fault prediction.

[0095] By constructing an initial line fault prediction model consisting of a deep convolutional neural network part and a support vector machine classification layer, the deep convolutional neural network can automatically extract multi-level features of the data, while the support vector machine has good performance in classification tasks. The combination of the two realizes the complementary advantages of feature extraction and classification functions. Compared with existing technologies, this model can process complex transmission line data more accurately, effectively improve the accuracy and reliability of fault prediction, better adapt to the diversity and complexity of transmission line operating conditions, and reduce the probability of false alarms and missed faults.

[0096] On the basis of the above embodiment, the model processing module trains the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed until the fireworks fitness value set meets the preset optimization conditions, and obtains the line fault prediction model, which is specifically used for:

[0097] According to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection result corresponding to each data to be analyzed, a sample set is constructed, wherein each sample in the sample set includes: the data to be analyzed collected during a fault prediction, and the actual fault detection result corresponding to this fault prediction;

[0098] Based on the sample set and the preset gradient optimization algorithm, a first initial prediction model in the initial line fault prediction model is trained to obtain a first prediction model, where the first prediction model is a model obtained by optimizing model parameters of the first initial prediction model based on the preset gradient optimization algorithm;

[0099] Determine, according to the first prediction model, a global feature corresponding to each sample in the sample set;

[0100] Based on the actual fault detection result of each sample in the sample set and the corresponding global feature, the second initial prediction model in the initial line fault prediction model is trained until the fireworks fitness value set meets the preset optimization condition, thereby obtaining the second prediction model;

[0101] A line fault prediction model is obtained according to the first prediction model and the second prediction model.

[0102] Among them, the preset historical time period can be determined based on the fault cycle, equipment maintenance cycle, line upgrade and renovation time, or data volume and data quality requirements. For example, it is necessary to ensure that the amount of data in the historical time period is rich enough to support the effective training of the model. If the amount of data is too small, the model may not be able to learn enough features and rules. When the transmission line is upgraded, its performance and operating characteristics will change. If the time of the last upgrade of the line is known, the historical time period can be determined from that point in time, combined with factors such as the speed of accumulation of line operation data and the frequency of faults.

[0103] The sample set may be basic data for allowing the model to learn the relationship between data features and failure outcomes.

[0104] The preset gradient optimization algorithm is a method for optimizing model parameters, such as the SGDM (Stochastic Gradient Descent with Momentum) algorithm, which gradually optimizes the model by calculating the gradient of the loss function with respect to the model parameters and updating the parameters in the opposite direction of the gradient. Here, the data in the sample set is used to input the first initial prediction model, and the loss is calculated based on the difference between the predicted output of the model and the actual fault detection result. Then, the preset gradient optimization algorithm is used to adjust and optimize the parameters in the first initial prediction model (such as the weights and biases in the convolution layer, pooling layer, and fully connected layer) to obtain the first prediction model, so that it can better extract data features and perform preliminary feature representation.

[0105] It should be noted that the traditional gradient descent algorithm may converge slowly when facing complex loss function terrain. Therefore, the preset gradient optimization algorithm can also be the Nesterov Accelerated Gradient (NAG) algorithm with momentum. This algorithm introduces a momentum term so that the model parameters can use the previous update information when updating, and have a certain inertia, so that it can move faster in areas where the gradient direction does not change much, accelerate the convergence speed of the model, and can obtain a better model faster, thereby improving the efficiency of fault prediction.

[0106] For example, 1) initialize all parameters in the first initial prediction model (based on the deep convolutional neural network part), including the convolution kernel weights and biases of the convolution layer, the parameters of the pooling layer (if there are trainable parameters), and the weights and biases of the fully connected layer. At the same time, set the hyperparameters of the NAG algorithm, such as the learning rate (usually between 0.001-0.1) and the momentum parameter β (generally around 0.9). 2) Input the sample data in the sample set into the first initial prediction model, and obtain the predicted output after operations such as convolution, pooling, and full connection. For example, predict whether a transmission line fails (the output is the probability of failure), and calculate the loss function based on the predicted output and the actual fault detection results corresponding to the sample. 3) First calculate the gradient under the current parameters, use the Nesterov acceleration technique, and calculate the "forward" gradient estimation point based on the weighted average of the gradient update direction of the previous iteration (i.e., the momentum term). Through this estimation point, the model can "see in advance" the direction of parameter update to a certain extent, making the update more effective. 4) Update the momentum term and model parameters based on the preset formula. 5) Repeat steps 2) to 4) until the training stop condition is met. The training stop condition may be reaching a predetermined number of iterations, the loss function converges to a smaller value, or the accuracy on the validation set no longer improves. Through continuous iterations, the parameters of the first initial prediction model are optimized, and finally the first prediction model is obtained, so that the performance of extracting the characteristics of the transmission line data is improved.

[0107] After parameter optimization and adjustment, the first prediction model can extract and output the global features corresponding to each sample from the input data. These global features integrate various local feature information in the data, better reflect the overall characteristics of the data, and provide more valuable input for the subsequent classification of the second initial prediction model.

[0108] The second initial prediction model is the support vector machine classification layer, which uses the actual fault detection results of each sample in the sample set and the corresponding global features output by the first prediction model to train the support vector machine. During the training process, the parameters of the support vector machine (such as kernel function and penalty factor) are optimized by the preset fireworks algorithm, and the optimization effect is measured by the fireworks fitness value set. The parameters are continuously adjusted until the preset optimization conditions are met, and the second prediction model is obtained, which can more accurately classify whether the transmission line has a fault according to the global features.

[0109] The final line fault prediction model is composed of the trained first prediction model and the second prediction model. The first prediction model is responsible for extracting the feature representation of the data, and the second prediction model is responsible for fault classification based on these features. The two work together to realize the prediction of transmission line faults.

[0110] It should be noted that the model training order here can be to first train the first prediction model separately, and when the performance of the first prediction model reaches a relatively stable state, apply it to the training sample to obtain output features. These output features are used as the input of the second prediction model, and then combined with the actual fault detection results in the training samples, the second prediction model is trained. This method is relatively simple and intuitive. The training process of the first prediction model is relatively independent, and its powerful automatic feature extraction capabilities can be fully utilized, focusing on the intrinsic structure and feature representation of the learning data. After the training of the first prediction model is stable, the second prediction model can be trained based on relatively fixed high-quality features, which is conducive to the second prediction model to better understand the feature space, thereby more effectively building a classification model.

[0111] However, since the training of the first prediction model is completed first, it may learn some features that are not necessarily the most critical in the subsequent classification of the second prediction model, resulting in more effort to adjust the parameters in the second prediction model training phase to adapt to these features, and may miss some better feature combinations that can be discovered through joint training. Therefore, the model training order can also refer to alternating the adjustment of the first prediction model and the second prediction model. For example:

[0112] First, input the training sample into the first prediction model to obtain the initial output features. Then input these features into the second prediction model, and calculate the initial loss of the second prediction model in combination with the actual fault detection results in the training samples. Next, perform alternating training. Each time the parameters of the first prediction model are adjusted, the updated first prediction model is immediately applied to the training sample to obtain new output features, and then these new features are used to adjust the parameters of the second prediction model. This process is repeated. Each time the parameters of the first prediction model or the second prediction model are adjusted, the loss is recalculated based on the new model output and actual results, and the corresponding parameters are updated until the performance of the entire line fault prediction model is optimized. Through alternating training, the first prediction model can adjust its feature extraction method in a timely manner according to the feedback of the second prediction model, so that the extracted features are more in line with the classification requirements of the second prediction model. At the same time, the second prediction model can also better adapt to the changing feature output of the first prediction model, thereby finding a better classification boundary.

[0113] By using different optimization algorithms to optimize the performance of the deep convolutional neural network part and the support vector machine part, the advantages of the two different model structures can be fully utilized, and the overall performance of the model can be gradually improved in a targeted manner during the training process. Compared with the existing technology, this staged, multi-algorithm optimization training method can improve the accuracy and stability of the model's prediction of transmission line faults, and better adapt to the complex operating data and variable fault modes of transmission lines.

[0114] On the basis of the above embodiment, the model processing module trains the second initial prediction model in the initial line fault prediction model based on the actual fault detection result of each sample in the sample set and the corresponding global feature until the fireworks fitness value set meets the preset optimization condition and the second prediction model is obtained, which is specifically used to:

[0115] Taking the parameters to be optimized of the second initial prediction model as optimization samples, and determining the number and value range of the optimization samples;

[0116] Determine the values ​​of multiple optimized samples according to the number and value range of the optimized samples;

[0117] For each optimized sample, the global feature corresponding to each sample in the sample set is input into the second initial prediction model corresponding to the optimized sample to obtain the fault prediction result corresponding to each sample in the sample set;

[0118] Determine a first fireworks fitness value of the optimized sample according to the actual fault detection result and the corresponding fault prediction result of each sample in the sample set, wherein the fireworks fitness value is used to indicate the accuracy of the second initial prediction model corresponding to the optimized sample in predicting the transmission line fault;

[0119] According to the first firework fitness value of each optimized sample, all optimized samples are subjected to explosion processing to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample;

[0120] For each optimized sample, each spark in the optimized sample is used as a new optimized sample, and the second firework fitness value of each new optimized sample is determined;

[0121] Determine a plurality of target optimized samples and a fireworks fitness value of each target optimized sample according to the first fireworks fitness value of each optimized sample and the second fireworks fitness value of each new optimized sample;

[0122] If the fireworks fitness values ​​of all target optimized samples meet the preset accuracy requirement, a final optimized sample is determined from all target optimized samples, wherein the model corresponding to the final optimized sample is the second prediction model.

[0123] In this content, first, the parameters that need to be adjusted and optimized in the model are regarded as optimization samples, and the number of these optimization samples (such as 500) and the range of values ​​of each parameter in the sample are determined. Then, the parameter values ​​in the optimization samples are randomly generated, such as the specific values ​​of the kernel function and the penalty factor in each optimization sample. Then, for the model represented by each determined optimization sample parameter, the global characteristics of all samples are input, and the model is run and the fault prediction of each sample is output. By comparing the actual fault situation with the model prediction result, the first fireworks fitness value is calculated to measure the model prediction accuracy corresponding to the group of optimization sample parameters. The higher the fitness value, the more accurate the model. After that, the mechanism of the fireworks algorithm is used to determine the explosion range (explosion radius) of each optimization sample (similar to fireworks) and the number of new samples (sparks) generated according to the fitness value, in order to conduct more exploration near the better sample parameters to find a better solution. The new samples (sparks) generated by the explosion are regarded as new optimization samples to be studied, and their fitness values ​​are calculated again to evaluate the accuracy of the model they represent. The fitness values ​​of the original optimized samples and the newly generated optimized samples are combined to screen out some better samples (target optimized samples) and their corresponding fitness values. Check whether the fitness values ​​of the selected target optimized samples meet the pre-set requirements for model accuracy. If they meet all requirements, an optimal (such as the one with the largest fireworks fitness value) optimized sample is determined as the final optimized sample, and its corresponding model is the trained second prediction model.

[0124] If there is a situation where the fireworks fitness values ​​of all target optimized samples do not meet the preset accuracy requirements, the target optimized sample is used as the optimized sample, and the steps of performing explosion processing on all optimized samples according to the first fireworks fitness value of each optimized sample are executed again to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample, until the fireworks fitness values ​​of all target optimized samples meet the preset accuracy requirements and / or the number of executions meets the preset number of iterations, and the final optimized sample is determined.

[0125] The fireworks explosion radius may refer to a range parameter determined according to the fitness value of the optimized sample in the fireworks algorithm, which is used to control the distribution range of new samples (sparks) generated by the optimized sample. The explosion radius of the optimized sample with a better fitness value is relatively small to facilitate a detailed search in its vicinity.

[0126] The number of sparks can refer to the number of new optimized samples generated by the optimized samples according to factors such as the explosion radius of their fireworks (similar to the sparks generated by the explosion of fireworks). Optimized samples with better fitness values may generate more sparks, that is, explore the nearby areas more. Optionally, to prevent the number of sparks generated by some fireworks from being too large or too small, the maximum and minimum values of the sparks can be set. When the number of generated sparks is greater than the maximum value, the current number of sparks is modified to the maximum value. When the number of generated sparks is less than the minimum value, the current number of sparks is modified to the minimum value.

[0127] In one example, the method for determining multiple target optimized samples according to the first fireworks fitness value of each optimized sample and the second fireworks fitness value of each new optimized sample can be to normalize the first fireworks fitness value and the second fireworks fitness value, and select multiple target optimized samples based on the normalized fireworks fitness values of all samples. Further, add up all the first fireworks fitness values and the second fireworks fitness values to obtain the sum of the fireworks fitness values of all optimized samples; for each optimized sample, divide its own fireworks fitness value by the sum to obtain the normalized fireworks fitness value, that is, the probability that the sample is selected; create an array with the same length as the number of remaining optimized samples to store the cumulative probability; starting from the first optimized sample, the cumulative probability of the first optimized sample is equal to its own normalized fireworks fitness value (probability), that is, P1 = p1. For the i-th (i>1) optimized sample, its cumulative probability is equal to the sum of the cumulative probabilities of the previous i - 1 optimized samples plus its own normalized fireworks fitness value (probability), that is, P i = P i-1 + p i , calculate the cumulative probability corresponding to each optimized sample in turn, and finally obtain the cumulative probability array [P1, P2,..., P n of all optimized samples (n optimized samples); use the random number generation function to generate a random number r in the range of 0 to 1; start from the starting point (the individual corresponding to the first element) of the cumulative probability array, and compare the cumulative probability with the random number one by one. If P1≥r, then the first optimized sample is the selected target optimized sample; if P1<r, continue to compare the size of P2 and r, and so on, until a cumulative probability P j greater than or equal to the random number r is found for the optimized sample. At this time, the j-th optimized sample is the target optimized sample selected by the roulette wheel method and will participate in the next iteration until the required number of target optimized samples is selected.

[0128] In another example, if there are 500 optimized samples and 300 new optimized samples, the method for determining multiple target optimized samples can be to sort the 800 samples according to their fireworks fitness values, and select samples with fireworks fitness values ​​higher than a preset value as target optimized samples; or, it can be to select samples that meet a preset number after sorting the fireworks fitness values ​​from high to low as target optimized samples, such as 350 optimized samples and 150 new optimized samples among the first 500 samples.

[0129] By taking the parameters to be optimized of the transmission line fault prediction model as optimization samples, multiple rounds of optimization are performed based on the sample set using the fireworks algorithm, including determining the optimization sample value, calculating the fitness value, generating new samples by explosion processing and re-evaluating, etc., the model parameters can be determined more accurately, and the accuracy of the model in predicting transmission line faults can be effectively improved. Compared with the existing technology, the prediction accuracy has been significantly improved, which can better meet the needs of transmission line fault prediction.

[0130] On the basis of the above embodiment, when the model processing module performs explosion processing on all optimized samples according to the first fireworks fitness value of each optimized sample to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample, it is specifically used to: perform explosion processing on all optimized samples according to the first fireworks fitness value of each optimized sample to obtain the initial fireworks explosion radius and the number of sparks corresponding to each optimized sample; based on the Tent chaotic mapping, adjust the range and distribution characteristics of the initial fireworks explosion radius to obtain the fireworks explosion radius.

[0131] Among them, explosion processing can refer to an operation under the framework of the fireworks algorithm, which is based on the optimization sample (analogous to fireworks), determines a range (initial fireworks explosion radius) according to its fitness value and other factors, and generates a certain number of new samples (sparks) within this range. The purpose is to explore the solution space where the model parameters are located in this way, and find a better combination of parameters to improve model performance. The initial fireworks explosion radius can be the range size indicator of the generation of new samples (sparks) corresponding to each optimized sample initially determined when performing the explosion processing operation based on the fitness value, and will be further adjusted through the Tent chaos map later.

[0132] Tent chaos mapping is a specific mathematical mapping method that can transform the input data according to its own unique chaos rules, so that the output data has chaotic characteristics, that is, the data distribution becomes complex and irregular and can be traversed relatively evenly within a certain range. It is used here to optimize the variation range and distribution characteristics of the initial fireworks explosion radius, and to assist in improving the ability to find the optimal model parameters.

[0133] By introducing Tent chaos mapping to further adjust the range and distribution characteristics of the fireworks explosion radius when performing explosion processing based on the fitness value of the optimized sample to determine the fireworks explosion radius and the number of sparks. Compared with the existing technology, this method can make the search range of model parameters more reasonable and cover the solution space more comprehensively, effectively avoid falling into the local optimal solution, and thus find the optimal model parameter combination more accurately.

[0134] On the basis of the above embodiment, the data analysis platform also includes a data preprocessing module, which is communicated with the data acquisition module and the model processing module respectively; the data preprocessing module is used to decompose the data to be analyzed in the data acquisition module based on wavelet transform to obtain decomposed data to be analyzed, and the decomposed data to be analyzed is characteristic data in the frequency domain; the data preprocessing module is also used to normalize the decomposed data to be analyzed to obtain preprocessed data to be analyzed, wherein the data to be analyzed in the model processing module is the preprocessed data to be analyzed.

[0135] Among them, wavelet transform can decompose the signal into components of different frequencies, has the characteristics of multi-resolution analysis, can analyze the data in the time and frequency domains simultaneously, and effectively extract the characteristic information in the data.

[0136] Normalization refers to the operation of scaling data according to a specific algorithm so that the data falls within a preset range (such as [0,1]). The purpose is to unify the magnitude and scale of the data to facilitate data comparison, analysis, and processing in the model, and to avoid deviations in the analysis results due to the numerical characteristics of the data itself.

[0137] By setting up a data preprocessing module in the data analysis platform, the wavelet transform is used to decompose the data to be analyzed to obtain frequency domain feature data, thereby better utilizing the model's analysis capabilities in the frequency domain. Compared with existing technologies, the wavelet transform can more accurately mine the feature information of the data to be analyzed at different frequencies and highlight the hidden features related to the fault, while the normalization process solves the problem of data magnitude differences, allowing the data to play a more balanced role in the model processing process, providing a high-quality data foundation for subsequent model processing.

[0138] On the basis of the above embodiment, when the data preprocessing module decomposes the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, it is specifically used to: perform three-layer decomposition processing on the data to be analyzed in the data acquisition module to determine the energy of the data to be analyzed in each frequency band and the total energy of all frequency bands; determine the energy proportion of each frequency band and the energy entropy corresponding to the energy proportion according to the energy of each frequency band and the total energy of all frequency bands; determine the decomposed data to be analyzed according to the energy entropy of each frequency band.

[0139] The three-layer decomposition process may refer to decomposing the original data to be analyzed into three layers of sub-data in different frequency bands according to different scales by using wavelet transform. Generally, as the number of decomposition layers increases, the frequency bands will be divided into finer ones, and the characteristics of the data in different frequency ranges can be analyzed more finely.

[0140] The energy of the data to be analyzed in each frequency band and the sum of the energy in all frequency bands can satisfy:

[0141]

[0142] Among them, E 3,j is the energy of a data to be analyzed in the jth frequency band, j = 1, 2, ..., 8; E0 is the sum of the energy of a data to be analyzed in all frequency bands; x j,a It represents the non-negative amplitude of the signal after reconstruction in the jth frequency band, and n represents the number of sampling points.

[0143] The energy proportion of each frequency band and the energy entropy corresponding to the energy proportion can satisfy:

[0144]

[0145] H=-∑p E logp E ;

[0146] Among them, p j is the energy proportion of each frequency band; H is the energy entropy corresponding to the energy proportion.

[0147] The decomposed data to be analyzed are characteristic data formed by the standard deviation energy entropy of the frequency band corresponding to the original data to be analyzed.

[0148] By calculating the energy, total energy, energy proportion and energy entropy of each frequency band, the characteristic distribution of data in the frequency domain can be analyzed more accurately. Compared with the existing technology, this method of determining the data to be analyzed after decomposition based on energy entropy can not only effectively extract the key frequency domain features in the data and highlight the important information related to the transmission line fault, but also achieve reasonable dimensionality reduction of data, reduce data redundancy, and provide more representative and high-quality data for subsequent model processing.

[0149] Based on the above embodiment, the data analysis platform also includes: a model evaluation module; the model evaluation module is used to evaluate the prediction performance of the line fault prediction model based on preset evaluation indicators after the model processing module obtains the line fault prediction model, and the preset evaluation indicators include the accuracy, precision, recall rate of the model prediction, and the harmonic mean of the precision and recall rate.

[0150] Among them, accuracy can refer to the ratio of the number of samples predicted correctly by the model to the total number of samples. It reflects the overall prediction accuracy of the model and satisfies: Acc = (TP+TN) / (TP+TN+FP+FN), where TP is the number of true positive examples, TN is the number of true negative examples, FP is the number of false positive examples, and FN is the number of false negative examples.

[0151] Precision refers to the ratio of the number of samples predicted by the model as positive samples (faulty samples) and actually positive samples to the number of samples predicted by the model as positive samples. It measures the accuracy of the model when predicting faults, and satisfies: Prec = TP / (TP+FP). For example, if the model predicts 30 faulty samples, of which 25 are actually faulty samples, then the precision is (25 / 30) × 100%.

[0152] The recall rate refers to the ratio of the number of samples predicted by the model as positive samples and actually positive samples to the number of actual positive samples, which reflects the coverage of the model for positive samples, that is, how many actual fault samples can be correctly identified, satisfying: Rec = TP / (TP+FN). For example, there are actually 40 fault samples, and the model predicts 30, of which 25 are correct, then the recall rate is (25 / 40) × 100%.

[0153] The harmonic mean of precision and recall is the F value, which takes precision and recall into consideration and can evaluate model performance in a more balanced way, satisfying the following equation: Acc = 2×(Prec×Rec) / (Prec+Rec).

[0154] By adding a model evaluation module to the data analysis platform, the performance of the line fault prediction model is evaluated using preset evaluation indicators such as accuracy, precision, recall, and F value, which can more comprehensively and objectively measure the performance of the model in the task of transmission line fault prediction. Not only can the correctness of the overall prediction of the model be accurately judged, but also the accuracy and coverage of the model's prediction of fault samples can be deeply analyzed, which helps to improve the prediction performance of the model.

[0155] Based on the contents described in the above embodiments, refer to Figure 2 , Figure 2 This is a module diagram of the data analysis platform provided in this application. The data analysis platform 20 provided in this embodiment includes:

[0156] The data acquisition module 201 is used to receive the data to be analyzed sent by the data acquisition terminal;

[0157] The data preprocessing module 202 is used to decompose the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, and the decomposed data to be analyzed is the characteristic data in the frequency domain;

[0158] The data preprocessing module 202 is further used to normalize the decomposed data to be analyzed to obtain preprocessed data to be analyzed, wherein the data to be analyzed in the model processing module is the preprocessed data to be analyzed;

[0159] The data processing module 203 is used to input the data to be analyzed into the line fault prediction model, obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal;

[0160] The model processing module 203 is further used to train the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed, until the fireworks fitness value set meets the preset optimization conditions, thereby obtaining the line fault prediction model;

[0161] The model evaluation module 204 is used to evaluate the prediction performance of the line fault prediction model based on preset evaluation indicators after the model processing module obtains the line fault prediction model.

[0162] Based on the above embodiment, the system also includes: a control module; the control module is used to perform fault processing operations on the transmission line based on the processing strategy generated by the data processing terminal; the control module is also used to receive processing instructions determined by maintenance personnel based on early warning information, and perform fault processing operations on the transmission line according to the processing instructions.

[0163] In this content, after a series of processing such as analyzing the data related to the transmission line and calculating the prediction model, the data processing terminal will generate a processing strategy for possible fault conditions. After the control module obtains these processing strategies, it performs corresponding operations according to the content of the strategy to handle the transmission line fault. For example, if the processing strategy is to cut off a specific branch of a line to isolate the fault area, the control module will send a control signal to the relevant power equipment (such as a circuit breaker, etc.) to make it perform the cutting operation. When the data analysis platform issues a transmission line fault warning information, the maintenance personnel determine the specific processing instructions based on their own experience, the actual situation on the site, and the detailed content of the warning information. The control module receives these instructions issued by the maintenance personnel, and then drives the corresponding power equipment or executes the relevant process to handle the fault. For example, if the maintenance personnel determines that a certain fault can be alleviated by temporarily adjusting the line load, they will issue the corresponding instructions, and the control module will coordinate the relevant equipment to perform load adjustment operations.

[0164] In some examples, the control module may be specifically configured to:

[0165] 1) Continuously monitor the information transmitted by the data processing terminal. When a high-priority emergency processing strategy is received, immediately parse the strategy content to determine the power equipment that needs to be operated (such as circuit breakers, switches, transformers, etc.) and their operation sequence.

[0166] 2) According to the analyzed operation sequence, control signals are sent to the corresponding power equipment to start the automated fault handling process. For example, a trip command is first sent to the circuit breaker of the faulty branch line, and then a closing command is sent to the switch of the backup line. At the same time, the fault detection equipment is started to collect data from the faulty area, and the collected data is sent back to the data analysis platform for further analysis of the fault cause and evaluation of the handling effect.

[0167] 3) When executing the medium priority strategy, the strategy content is also parsed, but some data feedback and verification links are added during the execution process. For example, when performing line load adjustment operations, the line current, voltage and other parameters are monitored in real time to ensure that the adjusted parameters meet the safety operation standards. If an abnormality is found, the operation is suspended and feedback is given to the data processing terminal, waiting for further instructions or making adjustments according to the preset abnormality handling rules.

[0168] 4) For the warning information corresponding to the low-priority observation strategy, it is only recorded and displayed, and no automatic processing operation is performed. However, the warning information is regularly checked to see whether it has been updated or upgraded (such as an increase in the probability of failure). If so, it is processed according to the corresponding priority strategy processing flow.

[0169] By setting up a control module in the system that can automatically handle faults based on the processing strategy generated by the data processing terminal and receive instructions from maintenance personnel to handle faults, an effective combination of automated processing and manual intervention is achieved. On the one hand, the data-driven processing strategy can quickly respond to common fault conditions, improve the efficiency of fault handling, and reduce power outage time; on the other hand, maintenance personnel can issue processing instructions based on early warning information, give full play to the advantages of manual experience in complex and special fault handling, and ensure the accuracy and reliability of fault handling, thereby improving the flexibility and effectiveness of the transmission line fault handling system, enhancing the ability of the power system to respond to faults, and ensuring the stability of power supply.

[0170] Based on the contents described in the above embodiments, the application scenario of this system can be to identify and predict whether wind deviation and leakage faults will occur during the operation of transmission lines in a wind farm environment (such as weak signals, unstable wind resources, frequent natural disasters, etc.) based on intelligent devices such as drones and sensors. Figure 3 A schematic diagram of the operation flow of an intelligent line fault prediction and identification system based on AI algorithm provided in this application is as follows: Figure 3 As shown, taking model training as an example, the operation process of the above-mentioned intelligent line fault prediction and identification system based on AI algorithm includes:

[0171] S301: Obtain a sensor data set.

[0172] Among them, sensors and other intelligent devices are deployed on the monitoring terminal of the wind farm a in place A, and the transmission line data and environmental data from February 1 to May 30 are selected. A total of 1,000 groups of data are intercepted for model training, each group contains data of 7 consecutive time units, and is divided into training set and test set in a ratio of 8:2. Optionally, the input data of each group includes current, voltage, power factor, wind speed, wind direction frequency, temperature, and humidity; the output data includes whether it is a wind deviation discharge fault data label.

[0173] S302, data preprocessing, including data cleaning, wavelet transform and data normalization.

[0174] Among them, data cleaning can remove any obvious errors or outliers. The db8 wavelet basis function with excellent time-frequency localization characteristics is used to perform three-layer wavelet packet decomposition on each signal data to obtain the feature set after the input data is layered decomposed in the frequency domain. The feature set data is normalized to the maximum and minimum values ​​to scale the data to the range of 0-1 to obtain the pre-processed data.

[0175] S303, inputting the preprocessed data (including wire data and environmental data) into the line fault prediction model composed of DC-CNN and SVM classification layers.

[0176] Among them, the DC-CNN model mainly includes 1 input layer (2 channels), 2×2 convolutional layers, 2×2 pooling layers, 2 fully connected layers and 1 output layer. The maximum number of DC-CNN training times is 2500, the initial learning rate is 0.001, the initial BatchSize is set to 16, and the initial value of Epochs is set to 20. After multiple rounds of parameter fine-tuning (the optimization algorithm is SGDM), it can be determined that the model performance is optimal when the learning rate is set to 0.00001, the Batch Size is set to 64, and the Epochs is set to 100, and the optimized DC-CNN model is obtained. Then the feature vector extracted by the fully connected layer is used as the input data of the SVM classification layer. Based on the improved fireworks algorithm, the SVM classification layer is trained to finally obtain the optimized line fault prediction model.

[0177] S304: Output the trained model fault recognition result.

[0178] In some embodiments, the above step S303 can be refined into the following steps:

[0179] Step 1: Structure and training of DC-CNN-SVM model:

[0180] 1) Input layer: The input preprocessed data is a 7×7 square matrix, and each row vector represents a set of data (seven variables) for one time unit.

[0181] 2) Convolutional layer: Satisfy

[0182] Among them, l is the number of convolutional layers; is the output; is the input signal; is the weight; is the bias; f(·) is the activation function ReLu; m j is the convolution range.

[0183] Both channels of the CNN network are two-layer structures. The convolution kernel size of the wire data channel is 5×5, and the number of kernels is 6; the convolution kernel size of the environment data channel is 3×3, and the number of kernels is 16, and the step size is set to 1×1.

[0184] 3) Pooling layer: Use the maximum pooling method to meet

[0185] in, is the multiplicative bias; is the additive bias, and down(·) is the maximum pooling function.

[0186] Both channels of the CNN network have a two-layer maximum pooling structure. The kernel size of the wire data channel is 5×1, and the number of cores is 6; the kernel size of the environment data channel is 3×1, and the number of cores is 16. The pooling window N is set to 2, and the step size is set to 1×2.

[0187] The multi-dimensional feature matrix obtained through the two channels is flattened into a one-dimensional feature vector (this can be done by adding a Flatten layer), and the new one-dimensional vector is input into the fully connected layer. It should be noted that the pooling layer has no parameters. The role of pooling is to retain the main features while reducing parameters and calculations to prevent overfitting.

[0188] 4) Fully connected layer: The obtained feature data is respectively passed through 128 and 64 neurons to learn global features, where the parameters of these neurons can be set based on experience. This layer satisfies:

[0189]

[0190] in, for Input; for and The weight between for ; M is the number of neurons in the B-1th layer; N is the number of neurons in the Bth layer.

[0191] 5) SVM classification layer: The feature vector extracted by the fully connected layer is used as the input data of the SVM classifier, and the label is used as the output (1 for yes and 0 for no).

[0192] Assume that the training sample is T = {(x i ,y i )}, i∈N, N is the number of samples, x i are the characteristics of the samples, y i is the category to which the sample belongs. Then solve a convex quadratic programming problem as shown below:

[0193]

[0194] Among them, C ≥ 0 is the penalty factor introduced to solve the problem of approximate linear separability, α i is the Lagrange multiplier. is a kernel function that can map low-dimensional data to high-dimensional data. i The value of , thus forming a classification decision function:

[0195] f(x)=sign(∑ i∈[1,N] y i α i K(x,x i )+b);

[0196] It can be found that the penalty factor and kernel function are two main parameters that directly affect the prediction effect of the network output. Since the circuit signal and environmental signal used are nonlinear signals, the RBF kernel function is more suitable than other kernel functions. The parameter optimization method of the SVM model uses the improved fireworks algorithm, selects accuracy as the evaluation index, optimizes the penalty coefficient and kernel function coefficient (the two are the positions of fireworks and sparks, respectively), and uses the five-fold cross-validation method for verification. The penalty coefficient can be finally determined to be 14 and the kernel function coefficient to be 0.0024.

[0197] The following is a detailed description of the process of improving the fireworks algorithm for model optimization.

[0198] 1) Initialization. Initialize a fireworks population of size N (500). Set the preset parameters of the fireworks algorithm, randomly initialize the fireworks population X according to the upper and lower bounds and dimensions of the fireworks particles (i.e., the initial parameter combination of SVM), and calculate the fitness of each firework according to the fitness calculation function, i.e., the DC-CNN-SVM model windage leakage fault accuracy rate. Assume that the dimension of the fireworks is D, and perform a displacement operation on the i-th firework in the k-th dimension, i.e.:

[0199]

[0200] Here, rand(1,-1) represents a uniform random number generated in the range (1,-1).

[0201] 2) Explosion. Calculate the explosion radius R of each firework explosion. i and the number of sparks S i , the calculation formula is as follows:

[0202]

[0203] Where R is the average explosion radius of fireworks; f i For fireworks x i The fitness function, y min and max is the extreme value of fitness in the population, ε is a small constant added to prevent the denominator from being zero, and m is a constant.

[0204] The variation range and distribution characteristics of the explosion radius are adjusted using Tent chaos mapping, and its mathematical expression is:

[0205]

[0206] Among them, x n is the value after the nth iteration, r is the system parameter, and its value range is 0 <r<1。

[0207] By adjusting the parameters r and iteration times of the Tent map, the ergodicity and distribution characteristics of the chaotic sequence can be controlled, thereby indirectly controlling the variation range and distribution characteristics of the explosion radius of the fireworks population. A larger r value and a higher number of iterations will make the chaotic sequence more violent, which may produce a larger explosion radius and make the fireworks population more dispersed; a smaller r value and a lower number of iterations will make the chaotic sequence more stable, produce a smaller explosion radius and make the fireworks population more concentrated.

[0208] In order to prevent some fireworks from generating too many or too few sparks, the number of sparks S i Subject to the following constraints:

[0209]

[0210] Among them, S min and S max They are the maximum and minimum values ​​of the number of sparks produced by a fireworks explosion.

[0211] 3) Gaussian mutation. Calculate the accuracy of windage discharge faults corresponding to model parameters of different sparks, select the spark with the highest accuracy to perform Gaussian mutation, and generate Gaussian mutation sparks. In order to improve the diversity of the firework particle population, randomly select a firework xi Perform Gaussian mutation operation in dimension k:

[0212]

[0213] Apply Gaussian mutation on the dimension k of the fireworks to transform the current fireworks x i Go to the fireworks with the best fitness x G The direction is adjusted, g is a random value that satisfies the Gaussian distribution with mean and variance of 1.

[0214] In order to ensure that the generated sparks are within the feasible domain, it is necessary to use specific rules to pull the sparks that may jump out of the feasible domain back into the feasible domain. The formula is as follows:

[0215]

[0216] in, and are the upper and lower boundaries of the feasible domain in dimension k respectively.

[0217] 4) Select the roulette strategy. Control all sparks (including sparks that have undergone Gaussian mutation and the next generation of fireworks) in the feasible domain space, use the roulette strategy to select the next generation of fireworks, pass the value to the next generation, and judge whether the optimal fitness value of the current set reaches the set ideal accuracy, or whether the number of iterations reaches the preset value. If the end condition is met, output the DC-CNN-SVM model parameters.

[0218] Optionally, keep the fireworks particles with the best fitness, and then normalize the fitness values. Normalize the fitness values ​​of all the individuals that are kept so that their sum is 1. This is to convert the fitness value into the probability of being selected. The normalized fitness value can be expressed as the individual fitness value divided by the sum of all individual fitness values.

[0219] Finally, select the fireworks particles that will participate in the next iteration, starting from the first individual among the retained fireworks particles, and accumulate the normalized fitness values ​​of each individual one by one to obtain the cumulative probability corresponding to each individual. The cumulative probability represents the total probability from the starting point of the roulette wheel to the end of the fan-shaped area where the current individual is located. Generate a random number between 0 and 1 to select a point on the roulette wheel. According to the generated random number, starting from the starting point of the cumulative probability array, compare the cumulative probability with the random number one by one until an individual with a cumulative probability greater than or equal to the random number is found. This individual is the selected individual, thereby obtaining a trained line fault prediction model. Optionally, the model performance can be evaluated using four indicators: selection accuracy, precision, recall rate, and F value to determine the prediction accuracy of the model.

[0220] The intelligent line fault prediction and identification system based on AI algorithm provided in the embodiment of the present application can use technologies such as big data and deep learning to realize automatic and intelligent identification and early warning, and improve the reliability and flexibility of the whole system. Intelligent monitoring equipment such as sensors and drones can collect the operation data of the whole system in real time, including parameters such as voltage, current, power, etc., input these parameters into the fault prediction model of DC-CNN-SVM, establish a remote monitoring platform, analyze the potential line fault risks through cloud computing and Internet of Things technology, issue early warning, and realize remote monitoring and management of the line system. For example, when the system predicts that an overload fault will occur soon, it will immediately cut off the power supply and issue an early warning message. Undoubtedly, the system plays a very important role in improving production efficiency and product quality, reducing maintenance costs and operating risks, and extending the service life of equipment.

[0221] Figure 4 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 4 As shown, the electronic device 40 provided in this embodiment includes: at least one processor 401 and a memory 402. Optionally, the device 40 also includes a communication component 403. The processor 401, the memory 402 and the communication component 403 are connected via a bus 404.

[0222] In the specific implementation process, at least one processor 401 executes the computer execution instructions stored in the memory 402, so that at least one processor 401 executes each step executed by the above system. For details, please refer to the relevant description in the above method embodiment.

[0223] The specific implementation process of the processor 401 can refer to the various steps executed by the above system, and specifically refer to the above method embodiment. Its implementation principle and technical effect are similar and will not be repeated here.

[0224] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be executed by a combination of hardware and software modules in the processor.

[0225] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (Non-volatile Memory, NVM), such as at least one disk memory.

[0226] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.

[0227] The present application also provides a computer program product, including a computer program, which implements the various steps executed by the above system when executed by a processor. For details, please refer to the relevant description in the above method embodiment.

[0228] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the various steps of the above-mentioned system execution are implemented. For details, please refer to the relevant description in the above-mentioned method embodiment.

[0229] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0230] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0231] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0232] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0233] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0234] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0235] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0236] Finally, it should be noted that those skilled in the art will readily conceive of other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses or adaptations of the present invention, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. An intelligent line fault prediction and identification system based on AI algorithm, characterized in that: include: Data collection terminals, data analysis platforms and data processing terminals that are interconnected and communicate with each other; The data acquisition terminal is used to collect the data to be analyzed of the transmission line according to the preset collection time requirement, and send it to the data analysis platform, wherein the data to be analyzed includes the power data of the transmission line and the environmental data of the location where the transmission line is located; The data analysis platform is used to receive the data to be analyzed, and input the data to be analyzed into the line fault prediction model to obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal; wherein the line fault prediction model is a model obtained by training a deep convolutional neural network and a support vector machine; The data processing terminal is used to receive the fault prediction result, and generate a processing strategy and early warning information for the transmission line based on the fault prediction result.

2. The system according to claim 1, characterized in that The data analysis platform includes a data acquisition module and a model processing module that are communicatively connected to each other; The data acquisition module is used to receive the data to be analyzed sent by the data acquisition terminal; The model processing module is used to input the data to be analyzed into the line fault prediction model, obtain the fault prediction result of the transmission line, and send the fault prediction result to the data processing terminal; The model processing module is also used to train the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed, until the fireworks fitness value set meets the preset optimization conditions, and the line fault prediction model is obtained; The initial line fault prediction model is constructed based on a deep convolutional neural network and a support vector machine, and the fireworks fitness value set is determined by optimizing the parameters in the initial line fault prediction model according to a preset fireworks algorithm.

3. The system according to claim 2, characterized in that The model processing module is also used for: Before training the pre-constructed initial line fault prediction model, construct an initial line fault prediction model, wherein the initial line fault prediction model includes a first initial prediction model consisting of an input layer, a convolution layer, a pooling layer, and a fully connected layer, and a second initial prediction model consisting of a support vector machine classification layer; Among them, in the first initial prediction model, the input layer is used to receive the data to be analyzed or the historical data to be analyzed, the convolution layer is used to extract the local features of the data in the input layer, the pooling layer is used to perform dimension conversion on the local features to obtain converted local features, and the fully connected layer is used to obtain global features according to the converted local features; In the second initial prediction model, the support vector machine classification layer is connected behind the fully connected layer in the first initial prediction model, and is used to receive the global features output by the first prediction model and classify the global features to predict whether a transmission line fault occurs.

4. The system according to claim 3, characterized in that The model processing module trains the pre-built initial line fault prediction model according to the data to be analyzed of the transmission line in a preset historical time period and the actual fault detection results corresponding to each data to be analyzed until the fireworks fitness value set meets the preset optimization conditions, and obtains the line fault prediction model, which is specifically used to: According to the data to be analyzed of the power transmission line in a preset historical time period and the actual fault detection result corresponding to each data to be analyzed, a sample set is constructed, wherein each sample in the sample set includes: the data to be analyzed collected during a fault prediction, and the actual fault detection result corresponding to this fault prediction; Based on the sample set and the preset gradient optimization algorithm, a first initial prediction model in the initial line fault prediction model is trained to obtain a first prediction model, wherein the first prediction model is a model obtained by optimizing model parameters of the first initial prediction model based on the preset gradient optimization algorithm; Determine, according to the first prediction model, a global feature corresponding to each sample in the sample set; Based on the actual fault detection result of each sample in the sample set and the corresponding global feature, a second initial prediction model in the initial line fault prediction model is trained until the fireworks fitness value set meets the preset optimization condition, thereby obtaining a second prediction model; A line fault prediction model is obtained according to the first prediction model and the second prediction model.

5. The system according to claim 4, characterized in that The model processing module trains the second initial prediction model in the initial line fault prediction model based on the actual fault detection result of each sample in the sample set and the corresponding global feature until the fireworks fitness value set meets the preset optimization condition and the second prediction model is obtained, specifically for: Taking the parameters to be optimized of the second initial prediction model as optimization samples, and determining the number and value range of the optimization samples; Determining the values ​​of the plurality of optimized samples according to the number and value range of the optimized samples; For each optimized sample, inputting the global feature corresponding to each sample in the sample set into the second initial prediction model corresponding to the optimized sample to obtain a fault prediction result corresponding to each sample in the sample set; Determine a first fireworks fitness value of the optimized sample according to an actual fault detection result and a corresponding fault prediction result of each sample in the sample set, wherein the fireworks fitness value is used to indicate the accuracy of the second initial prediction model corresponding to the optimized sample in predicting the transmission line fault; According to the first firework fitness value of each optimized sample, all optimized samples are subjected to explosion processing to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample; For each optimized sample, each spark in the optimized sample is used as a new optimized sample, and a second firework fitness value of each new optimized sample is determined; Determine a plurality of target optimized samples and a fireworks fitness value of each target optimized sample according to the first fireworks fitness value of each optimized sample and the second fireworks fitness value of each new optimized sample; If the fireworks fitness values ​​of all target optimized samples meet the preset accuracy requirement, a final optimized sample is determined from all target optimized samples, wherein the model corresponding to the final optimized sample is the second prediction model.

6. The system according to claim 5, characterized in that The model processing module performs explosion processing on all optimized samples according to the first fireworks fitness value of each optimized sample to obtain the fireworks explosion radius and the number of sparks corresponding to each optimized sample, specifically for: According to the first firework fitness value of each optimized sample, all optimized samples are subjected to explosion processing to obtain the initial firework explosion radius and number of sparks corresponding to each optimized sample; Based on the Tent chaotic map, the range and distribution characteristics of the initial fireworks explosion radius are adjusted to obtain the fireworks explosion radius.

7. The system according to claim 2, characterized in that The data analysis platform also includes a data preprocessing module, which is communicated with the data acquisition module and the model processing module respectively; The data preprocessing module is used to decompose the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, and the decomposed data to be analyzed is the characteristic data in the frequency domain; The data preprocessing module is further used to normalize the decomposed data to be analyzed to obtain preprocessed data to be analyzed, wherein the data to be analyzed in the model processing module is the preprocessed data to be analyzed.

8. The system according to claim 7, characterized in that When the data preprocessing module decomposes the data to be analyzed in the data acquisition module based on wavelet transform to obtain the decomposed data to be analyzed, it is specifically used to: Performing three-layer decomposition processing on the data to be analyzed in the data acquisition module to determine the energy of the data to be analyzed in each frequency band and the total energy in all frequency bands; Determine, according to the energy of each frequency band and the sum of the energies in all frequency bands, the energy proportion of each frequency band and the energy entropy corresponding to the energy proportion; According to the energy entropy of each frequency band, the decomposed data to be analyzed is determined.

9. The system according to any one of claims 2 to 8, characterized in that: The data analysis platform also includes: a model evaluation module; The model evaluation module is used to evaluate the prediction performance of the line fault prediction model based on preset evaluation indicators after the model processing module obtains the line fault prediction model. The preset evaluation indicators include the accuracy, precision, recall rate, and harmonic mean of the precision and recall rate of the model prediction.

10. The system according to any one of claims 1 to 8, characterized in that: The system further comprises: a control module; The control module is used to perform fault processing operations on the power transmission line based on the processing strategy generated by the data processing terminal; The control module is also used to receive a processing instruction determined by a maintenance person based on the warning information, and perform a fault processing operation on the power transmission line according to the processing instruction.

Images