A wind farm power collection line fault early warning method, system, device and medium

Abstract

This application relates to a method, system, equipment, and medium for early warning of faults in wind farm collector lines, belonging to the field of line fault detection technology in smart grids. The method includes: dividing the 35kV collector lines of the wind farm into multiple sections; determining whether a fault exists in each section; if so, identifying it as a faulty section; if not, dynamically calculating the deviation between the current environmental data and the environmental baseline data of the section to obtain a dynamic environmental feature vector; calculating the predicted dust accumulation growth rate and dust accumulation assessment index; identifying sections whose dust accumulation assessment index exceeds a preset threshold as dust accumulation risk sections; fusing the multi-dimensional feature vectors of the dust accumulation risk sections to generate a comprehensive fault feature vector and predicting latent arc faults, obtaining the risk probability of latent arc faults and determining the risk warning level, thereby generating risk warning information and risk maintenance strategies. This application can achieve rapid location of visible faults in collector lines and accurate prediction of latent faults.

Description

Technical Field

[0001] This application relates to the field of line fault detection technology for smart grids, and in particular to a method, system, equipment and medium for early warning of faults in wind farm collector lines. Background Technology

[0002] As a crucial component of renewable energy, the safety and stability of the core infrastructure—the collector lines—directly impact the power generation efficiency of wind farms and the stable operation of the power grid. With the development of smart grids and intelligent distribution systems, 35kV collector lines have gradually become an important research subject for the intelligent operation and maintenance of wind farms. As one of the most common transmission lines in wind farms, they operate within the medium-voltage transmission range, undertaking the task of transmitting large amounts of electrical energy while simultaneously meeting the dual requirements of stability and safety. However, due to their frequent distribution in harsh geographical environments, collector lines are susceptible to the combined effects of external environmental factors and operating conditions, leading to either overt or covert faults.

[0003] Currently, the problem of latent arcing faults in 35kV collector lines is particularly prominent in dusty environments. Because the line insulators are exposed to wind and sand, the dust accumulation gradually increases. When humidity changes or wind speeds increase sharply, the accumulated dust can cause abnormal increases in leakage current or local electric field distortion, leading to latent arcing faults. These faults typically initially manifest as weak, localized discharges. Due to their weak intensity and short duration, they are often not detected promptly using conventional threshold comparison methods. However, after prolonged accumulation, they can lead to serious problems such as insulator damage and line breakdown, gradually developing into overt faults that severely impact the operational safety of wind farms.

[0004] Therefore, given the complex operating scenarios of 35KV wind farm collector lines in the fields of smart grids and smart distribution, how to combine the complex operating environment of the collector lines to achieve rapid location of explicit faults and accurate prediction of implicit faults is an urgent problem to be solved. Summary of the Invention

[0005] To achieve rapid location of visible faults and accurate prediction of hidden faults, this application provides a method, system, equipment, and medium for early warning of faults in wind farm collector lines.

[0006] Firstly, this application provides a method for early warning of faults in wind farm collection lines, employing the following technical solution:

[0007] A method for early warning of faults in wind farm collector lines, the method comprising:

[0008] The wind farm's power collection line is divided into multiple sections according to the preset section division information;

[0009] Real-time acquisition of current monitoring data for each section; the current monitoring data includes current electrical parameters, current environmental data, and current dust accumulation thickness data;

[0010] Based on the current electrical parameters, determine whether there is a fault in each section; if so, identify the section as a faulty section and send a fault maintenance prompt message to the management terminal.

[0011] If not, the dynamic deviation between the current environmental data of the segment and the environmental baseline data is calculated to obtain the dynamic environmental feature vector of the segment.

[0012] Based on the dynamic environmental feature vector, the predicted dust accumulation growth rate of the section is calculated using a preset algorithm.

[0013] Calculate the dust accumulation assessment index of the section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate of the section;

[0014] The section where the dust accumulation assessment index exceeds a preset threshold is identified as a dust accumulation risk section;

[0015] The current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vector of the dust accumulation risk zone are fused to generate a comprehensive fault feature vector.

[0016] The comprehensive fault feature vector of the dust accumulation risk section is input into the pre-constructed hidden arc fault prediction model to obtain the hidden arc fault risk probability of the dust accumulation risk section.

[0017] Based on the probability of hidden arc fault risk, determine the risk warning level corresponding to the dust accumulation risk section;

[0018] Based on the risk warning level, generate risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section.

[0019] By adopting the above technical solutions, based on segmented management, multi-dimensional feature fusion, and neural network prediction, rapid location of visible faults in the collector lines of 35KV wind farms and accurate early warning and graded response to hidden faults have been achieved. By screening the electrical parameters of each segment for faults, segments with visible faults are quickly identified. In the absence of visible faults, the dust accumulation growth rate is dynamically predicted and the dust accumulation risk is assessed. By inputting multi-dimensional data into the neural network model, the probability and corresponding level of hidden arc fault risk are accurately output, thus determining the hidden fault segments. The final maintenance strategy optimizes resource scheduling based on the risk level, ensuring precise and efficient maintenance and significantly improving the safety, reliability, and intelligent operation and maintenance capabilities of the wind farm.

[0020] Optionally, the formula for calculating the predicted dust accumulation growth rate of the section is:

[0021]

[0022] In the above formula, ΔV, ΔH, and ΔT represent wind speed deviation, humidity deviation, and temperature deviation, respectively. 基线 H 基线 T 基线 These are the wind speed baseline, humidity baseline, and temperature baseline, respectively. 当前湿度 The current humidity data is given, and γ1, γ2, γ3, and γ4 are empirical adjustment coefficients, respectively.

[0023] Optionally, the system also includes a training step for the latent arc fault prediction model, the training step comprising:

[0024] Acquire historical datasets and perform preprocessing; the historical datasets include historical monitoring data and pre-labeled latent arc fault status tags, and the historical monitoring data includes historical dust accumulation thickness data, historical electrical parameters, historical dust accumulation growth rate and historical environmental data;

[0025] The historical monitoring data are fused with feature vectors to generate a sample feature set, which is then associated with the latent arc fault status label.

[0026] The sample feature set after label association is divided into a training set and a test set;

[0027] The sample feature set in the training set is input into a pre-constructed multi-layer feedforward neural network model to obtain training prediction values;

[0028] The training prediction values ​​are compared with the latent arc fault state labels in the training set, and the cross-entropy loss function is calculated.

[0029] The gradient is calculated and the model parameters are updated by backpropagation until the cross-entropy loss function meets the preset requirements, thus obtaining the trained implicit arc fault prediction model.

[0030] By adopting the above technical solution, based on feature fusion and deep learning models, complex correlations between multi-dimensional features can be extracted, accurate fault risk probabilities can be output, and model parameters can be dynamically updated using cross-entropy loss and backpropagation algorithms to improve training efficiency. Ultimately, a highly efficient and accurate prediction model is obtained, providing decision support for intelligent operation and maintenance and resource optimization. This technical solution significantly improves the safety and intelligence level of 35kV collector lines in wind farms.

[0031] Optionally, after obtaining the trained implicit arc fault prediction model, the method further includes:

[0032] The sample feature set in the test set is input into the trained latent arc fault prediction model to obtain the test prediction value;

[0033] The test prediction value is compared with the latent arc fault status label in the test set to obtain the test comparison result;

[0034] Based on the test comparison results, the performance index of the trained latent arc fault prediction model is evaluated and iterated until the preset performance index is met or the preset number of iterations is reached, thus obtaining the trained latent arc fault prediction model.

[0035] By adopting the above technical solutions, when performance indicators fail to meet the standards, the model structure and hyperparameters are adjusted for optimization, ensuring the reliability and generalization ability of the final model. The early stopping mechanism effectively avoids overfitting during model training, ensuring optimal performance on test data. Ultimately, the trained latent arc fault prediction model exhibits high accuracy and stability, enabling real-time and efficient early warning of latent faults in wind farm collector lines. This provides reliable guidance for the operation and maintenance team, improving the overall operational safety and efficiency of the wind farm.

[0036] Optionally, after the step of generating risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section based on the risk warning level, the method further includes:

[0037] Based on the risk warning level, the maintenance priority of the dust accumulation risk area is sorted to obtain a maintenance task list including multiple priority groups;

[0038] Determine whether there are multiple dust accumulation risk zones in each priority group;

[0039] If so, the dust risk zones in the same priority group are then sorted in a secondary order according to the dust accumulation assessment index to obtain a secondary order list of the priority group.

[0040] The maintenance task list is updated according to the secondary sorting list of each priority group;

[0041] Based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk area, maintenance task information is generated and sent to the maintenance terminal.

[0042] By adopting the above technical solutions, scientific maintenance and management of dust accumulation risk areas in wind farms has been achieved. Through a dual ranking system based on risk warning levels and dust accumulation assessment indices, the priority and execution sequence of maintenance tasks have been refined. Furthermore, dynamic maintenance strategies are used to generate maintenance task information, providing precise guidance for operation and maintenance personnel. The optimization and real-time updating of the task list not only improves resource utilization efficiency but also significantly reduces the risk of latent fault propagation.

[0043] Optionally, after determining the risk warning level corresponding to the dust accumulation risk section based on the probability of the latent arc fault risk, the method further includes:

[0044] Determine whether the risk warning level corresponding to the dust accumulation risk section exceeds the preset level threshold. If so, the dust accumulation risk section is determined as a high-risk section.

[0045] The topology of the wind farm's power collection lines is obtained based on the preset segment division information;

[0046] Based on the line topology, adjacent segments of the high-risk segment are identified, and an adjacent segment list is constructed; the adjacent segment list includes directly adjacent segments and indirectly adjacent segments;

[0047] Based on the dust accumulation assessment index, the similarity of dust accumulation characteristics between each adjacent segment and the high-risk segment is calculated.

[0048] Based on a preset similarity threshold, highly similar segments of the high-risk segments are determined according to the similarity of the dust accumulation features;

[0049] Historical electrical parameters of the high-risk section are obtained and preprocessed to obtain time series data of the high-risk section;

[0050] Time series modeling is performed based on the time series data, and residual analysis and anomaly detection are conducted based on the ARIMA model to determine the time points of abnormal fluctuations in the high-risk segment.

[0051] Obtain the historical electrical parameters of the highly similar section at the time point of the abnormal fluctuation and perform preprocessing;

[0052] Based on the abnormal fluctuation time points, electrical parameter synchronization analysis is performed on the high-risk section and the highly similar section to obtain the synchronization analysis results;

[0053] Based on the results of the synchronization analysis, it is determined whether there is synchronous fluctuation of electrical parameters. If so, the highly similar segments are identified as highly correlated segments, and a list of highly correlated segments is obtained.

[0054] By adopting the above technical solutions, multi-level and multi-dimensional fault propagation risk analysis was achieved. This not only provides early warning of faults in single sections but also reveals potential risks of fault propagation, providing a scientific basis for the intelligent maintenance of wind farm collector lines. Through real-time data-driven and intelligent analysis, this solution significantly improves the accuracy of fault management, reduces reliance on manual intervention, and promotes the intelligent and efficient operation and maintenance of wind farms.

[0055] Optionally, when the dust accumulation risk zone is a high-risk zone, the step of generating maintenance task information based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk zone includes:

[0056] The list of highly correlated zones is determined based on the dust accumulation risk zones;

[0057] Synchronous maintenance tasks are generated based on the list of highly correlated sections and added to the maintenance strategy of the dust accumulation risk section to obtain the updated maintenance strategy.

[0058] Based on the updated maintenance task list and the updated maintenance strategy, maintenance task information for the dust accumulation risk area is generated.

[0059] By adopting the above technical solutions, the lack of a holistic perspective in traditional wind farm maintenance was successfully addressed. By extending the maintenance of high-risk sections to related sections, potential fault propagation was prevented in a systematic way, significantly improving the operation and maintenance efficiency and safety of wind farms. The complete closed loop from fault identification to global maintenance effectively enhances the reliability and intelligent operation and maintenance level of wind farm collection lines.

[0060] Secondly, this application provides a fault early warning system for wind farm collection lines, which adopts the following technical solution:

[0061] A fault early warning system for wind farm collector lines, the system comprising:

[0062] The segment division module is used to divide the wind farm collection line into multiple segments according to preset segment division information;

[0063] The monitoring module is used to acquire the current monitoring data of each section in real time; the current monitoring data includes current electrical parameters, current environmental data, and current dust accumulation thickness data.

[0064] The judgment module is used to determine whether there is a fault in each section based on the current electrical parameters; if so, it outputs a first judgment result; if not, it outputs a second judgment result.

[0065] The fault indication module is used to determine the segment as a faulty segment in response to the first judgment result and send fault maintenance prompt information to the management terminal.

[0066] The dynamic deviation calculation module is used to perform dynamic deviation calculation between the current environmental data of the section and the environmental baseline data to obtain the dynamic environmental feature vector of the section.

[0067] The dust growth prediction module is used to combine the dynamic environmental feature vector and calculate the predicted dust growth rate of the section based on a preset algorithm.

[0068] The dust accumulation assessment module is used to calculate the dust accumulation assessment index of the section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate of the section.

[0069] The dust accumulation risk zone determination module is used to determine the zone where the dust accumulation assessment index exceeds a preset threshold as a dust accumulation risk zone.

[0070] The feature vector fusion module is used to fuse the current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vectors of the dust accumulation risk section to generate a comprehensive fault feature vector.

[0071] The latent arc fault prediction module is used to input the comprehensive fault feature vector of the dust accumulation risk section into the pre-constructed latent arc fault prediction model to obtain the latent arc fault risk probability of the dust accumulation risk section.

[0072] The early warning level determination module is used to determine the risk early warning level corresponding to the dust accumulation risk section based on the probability of the hidden arc fault risk.

[0073] The risk warning and maintenance module is used to generate risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section based on the risk warning level.

[0074] Thirdly, this application provides a computer device, which adopts the following technical solution:

[0075] A computer device includes a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to perform the steps of the method as described in the first aspect.

[0076] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution:

[0077] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as in any of the methods in the first aspect.

[0078] In summary, this application includes at least one of the following beneficial technical effects: it significantly improves the intelligence and precision of wind farm operation and maintenance, reduces the detection blind spots of hidden faults, effectively prevents the spread of potential faults, ensures the safe and efficient operation of wind farms, and optimizes resource allocation and maintenance costs. Attached Figure Description

[0079] Figure 1 This is a schematic diagram of the first process of a wind farm collector line fault early warning method according to an embodiment of this application.

[0080] Figure 2 This is a second flowchart illustrating a wind farm collector line fault early warning method according to an embodiment of this application.

[0081] Figure 3 This is a schematic diagram of the third process of a wind farm collector line fault early warning method according to an embodiment of this application.

[0082] Figure 4 This is a schematic diagram of the fourth process of a wind farm collector line fault early warning method according to an embodiment of this application.

[0083] Figure 5 This is a fifth flowchart of a wind farm collector line fault early warning method according to an embodiment of this application.

[0084] Figure 6 This is a sixth flowchart of a wind farm collector line fault early warning method according to an embodiment of this application. Detailed Implementation

[0085] To make the purpose, technical solution, and advantages of this application clearer, the following description is provided in conjunction with the appendix. Figure 1-6 The present application will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the application.

[0086] This application discloses a method for early warning of faults in the collector lines of a wind farm.

[0087] Reference Figure 1 A method for early warning of faults in wind farm collector lines, the method comprising:

[0088] Step S101: Divide the wind farm collection line into multiple sections according to the preset section division information;

[0089] Based on Geographic Information System (GIS) data, historical fault distribution, and environmental conditions, the 35kV collector lines of wind farms can be pre-divided into multiple sections to achieve differentiated management. This division can reflect the differences in dust accumulation, electrical characteristics, and environmental conditions among different sections, making subsequent monitoring and assessment more accurate.

[0090] In some embodiments, GIS data can be used to extract the terrain features of the line, including information such as the location of corner towers, high ground and low ground. By combining historical fault records, high-risk sections, such as sections with severe dust accumulation or frequent electrical faults, can be marked to obtain preset section division information.

[0091] Step S102: Acquire the current monitoring data of each section in real time; wherein, the current monitoring data includes the current electrical parameters, current environmental data, and current dust accumulation thickness data;

[0092] Specifically, sensor devices are attached to each section of the power line, including dust sensors, electrical parameter sensors (such as leakage current and electric field sensors), and environmental monitoring devices (such as wind speed, humidity, and temperature sensors). By collecting multi-dimensional data in real time, the operating status of the line and environmental change trends can be reflected. Among them, electrical parameters are directly related to line faults, while environmental data and dust thickness data indirectly reflect the risk of hidden faults.

[0093] Understandably, leakage current sensors measure the current intensity and fluctuation characteristics of each section, electric field sensors measure the local electric field intensity and distribution changes; wind speed, humidity and temperature sensors obtain the intensity of sand and dust activity, the conductivity of wet sand and dust and the influence of temperature changes on dust adhesion; and optical sensors use reflectivity changes to measure the dust thickness on the insulator surface.

[0094] Step S103: Determine whether there is a fault in each section based on the current electrical parameters; if yes, proceed to step S104; if no, proceed to step S105.

[0095] Step S104: Identify the section as the faulty section and send a fault maintenance prompt message to the management terminal;

[0096] Among them, electrical parameters (such as leakage current and electric field strength) are direct manifestations of obvious faults (such as short circuits and flashovers). By comparing them with preset thresholds, faulty sections can be quickly identified.

[0097] In one embodiment of this application, by comparing whether the leakage current value exceeds a safety threshold, a significant increase in leakage current may indicate insulation breakdown or a short circuit fault; by detecting whether the electric field distribution distortion exceeds a threshold, a localized strong electric field may indicate severe dust accumulation on the insulator or that a discharge has occurred. If any of the above fault conditions are met, the section is immediately marked as faulty and a maintenance prompt is triggered.

[0098] Understandably, using electrical parameters as the primary criterion for judgment can quickly screen for obvious faults, reduce the burden of complex calculations, and improve system response efficiency.

[0099] Step S105: Calculate the dynamic deviation between the current environmental data of the section and the environmental baseline data to obtain the dynamic environmental feature vector of the section.

[0100] Among them, dynamic deviation calculation can quantify the difference between the current environment and the baseline environment, capture changes in the intensity of dust activity, and provide a basis for calculating the dust accumulation growth rate.

[0101] Specifically, the deviation between the current environmental data (wind speed, temperature, humidity) and the environmental baseline data (which can be pre-configured based on historical data) is calculated to obtain the dynamic environmental feature vector:

[0102] F 环境 ={ΔV,ΔH,ΔT};

[0103] In the above formula, ΔV, ΔH, and ΔT represent wind speed deviation, humidity deviation, and temperature deviation, respectively.

[0104] Step S106: Combine the dynamic environmental feature vector and calculate the predicted dust growth rate of the section based on the preset algorithm.

[0105] Among them, the dust growth rate is an important indicator for measuring the dust accumulation trend on the surface of insulators. By combining environmental deviations and the current dust thickness, the accumulation rate of dust can be predicted.

[0106] Specifically, based on the dynamic environmental feature vector, a nonlinear function can be used to fit the predicted dust accumulation growth rate as: G 积尘 =f(ΔV,ΔH,ΔT).

[0107] In the above formula, G 积尘 The predicted dust accumulation rate is represented by , and f represents the mapping relationship between environmental characteristics and dust accumulation rate. Additionally, the effect of humidity on the conductivity of dust accumulation needs to be considered, and the model coefficients should be dynamically adjusted.

[0108] In one embodiment of this application, the specific formula for calculating the predicted dust growth rate is as follows:

[0109] ;

[0110] In the above formula, ΔV, ΔH, and ΔT represent wind speed deviation, humidity deviation, and temperature deviation, respectively. 基线 H 基线 T 基线 These are the wind speed baseline, humidity baseline, and temperature baseline, respectively. 当前湿度 The current humidity data is used because humidity directly affects the conductivity of dust accumulation; γ1, γ2, γ3, and γ4 are empirical adjustment coefficients that can be obtained through training with historical dust accumulation data.

[0111] Step S107: Calculate the dust accumulation assessment index of the section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate of the section.

[0112] Among them, a dust accumulation index model is constructed using the current dust accumulation thickness and the predicted dust accumulation growth rate as indicators, for example:

[0113] ;

[0114] Among them, D 当前 Given the current dust accumulation thickness, D 平均 G represents the average dust accumulation thickness in each section. 积尘G is the dust accumulation growth rate obtained by fitting a nonlinear function. 平均 The average historical dust accumulation growth rate is given by α and β, which are weighting coefficients (usually determined through experimental or training data), and α + β = 1.

[0115] Step S108: Determine the sections where the dust accumulation assessment index exceeds the preset threshold as dust accumulation risk sections;

[0116] Among them, the preset threshold can be pre-configured and adjusted according to the actual situation; through the dust accumulation assessment index, the possibility of hidden faults is quantified, and potential risks are quantified and specified, providing a quantitative basis for the identification and marking of dust accumulation risk areas.

[0117] Understandably, when there are no obvious faults, the thickness of dust accumulation is an important parameter affecting the performance of insulators. Combining the current dust accumulation thickness data with the predicted dust accumulation growth rate obtained through dynamic environmental characteristics can reflect the future fault risk and thus determine the dust accumulation risk section.

[0118] Step S109: The current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vector of the dust accumulation risk section are fused to generate a comprehensive fault feature vector.

[0119] The comprehensive fault feature vector is: X 故障 ={D 当前 G 积尘 ,I 泄漏 E 畸变 ,F 环境};

[0120] In the above formula, the current dust accumulation thickness D 当前 It can directly reflect the current dust accumulation status on the insulator surface, and the predicted dust growth rate G 积尘 The leakage current I of the current electrical parameters can quantify future dust accumulation trends. 泄漏 and electric field distortion intensity E 畸变 The dynamic environmental feature vector F can be used to characterize the risk level of current electrical anomalies. 环境 It can be used to assess the combined impact of the current environment on dust accumulation and electrical parameters.

[0121] It should be noted that, to ensure that data of different dimensions can be input into the model, each feature needs to be normalized (e.g., mapping the data to the [0,1] interval) to facilitate model processing. The fused comprehensive fault feature vector fully describes the risk characteristics of the segment and serves as the basic input for predicting latent faults.

[0122] Step S110: Input the comprehensive fault feature vector of the dust accumulation risk section into the pre-constructed latent arc fault prediction model to obtain the latent arc fault risk probability of the dust accumulation risk section.

[0123] Among them, latent arc faults refer to the weak, discontinuous arcs that occur on the insulation surface due to insulator performance degradation or local dust accumulation, but do not reach the level of complete breakdown. Its characteristics are: the arc occurs locally and lasts for a short time, and may not be able to trigger the line protection device immediately; these arcs can lead to local overheating or further aging of the insulator, making it more susceptible to serious faults.

[0124] Specifically, latent arc faults typically exhibit specific nonlinear correlations (such as when the dust accumulation thickness is high, electrical parameters are more prone to abnormal fluctuations). These complex nonlinear relationships can be extracted through neural network model training, enabling accurate prediction of fault risks.

[0125] It should be noted that the probability P of the hidden arc fault risk obtained by the model prediction 隐性故障 It is a continuous value (such as 0 to 1) used to indicate the probability of latent arc faults, which can provide a basis for subsequent risk level classification.

[0126] Step S111: Determine the risk warning level corresponding to the dust accumulation risk section based on the probability of latent arc fault risk;

[0127] The risk warning level classification is based on the numerical range of the failure risk probability. By converting continuous risk probabilities into discrete level descriptions, it facilitates management and maintenance. Specifically, risk warning levels are typically divided into low, medium, and high to reflect the severity of failure risks and guide subsequent maintenance strategies.

[0128] Step S112: Generate risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section based on the risk warning level.

[0129] Different risk levels correspond to different risk warning information and risk maintenance strategies to ensure efficient use of resources. For example, low-risk sections do not require immediate handling to reduce resource waste; medium-risk sections require key monitoring or cleaning to prevent latent faults from becoming apparent faults; and high-risk sections require immediate emergency maintenance measures to avoid threatening the safety of line operation.

[0130] In the above implementation, based on segmented management, multi-dimensional feature fusion, and neural network prediction, rapid location of visible faults in the 35KV wind farm's collector lines and accurate early warning and graded response to hidden faults are achieved. By screening the electrical parameters of each segment for faults, visible fault segments are quickly identified. In the absence of visible faults, the dust accumulation growth rate is dynamically predicted and the dust accumulation risk is assessed. By inputting multi-dimensional data into the neural network model, the probability and corresponding level of hidden arc fault risk are accurately output, thus determining the hidden fault segments. The final maintenance strategy optimizes resource scheduling based on the risk level, ensuring precise and efficient maintenance and significantly improving the safety, reliability, and intelligent operation and maintenance capabilities of the wind farm.

[0131] Reference Figure 2 As one implementation method for predicting latent arc faults, the training steps include:

[0132] Step S201: Obtain historical datasets and perform preprocessing;

[0133] The historical dataset includes historical monitoring data and pre-labeled implicit arc fault status tags. The historical monitoring data includes historical dust accumulation thickness data, historical electrical parameters, historical dust accumulation growth rate, and historical environmental data.

[0134] Specifically, the quality of historical datasets directly determines the accuracy and reliability of model training. Historical data collected from actual wind farms provides ample samples for the model, covering various operating states (such as normal operation and latent arc fault states). Data preprocessing steps include data cleaning and alignment, eliminating abnormal noise by handling outliers, while improving data consistency and reliability.

[0135] In some embodiments, the latent arc fault status label is used to indicate whether a sample has a latent fault. A binary label can be used, where 0 indicates that the sample has not experienced a latent arc fault in the historical data, and 1 indicates that the sample has indeed experienced a latent arc fault in the historical data.

[0136] Step S202: The historical monitoring data is fused with feature vectors to generate a sample feature set and associated with the latent arc fault status label.

[0137] Feature vector fusion aims to combine data from multiple dimensions (dust accumulation, electrical parameters, environmental characteristics, etc.) into a multi-dimensional feature vector, providing a complete description of the line's operating status. Each sample's feature vector is associated with its corresponding label, forming a sample data pair.

[0138] Step S203: Divide the sample feature set after label association into a training set and a test set;

[0139] Typically, the samples are divided into a training set (70%-80%) and a test set (20%-30%) to ensure that the model does not directly access the test data during training.

[0140] Step S204: Input the sample feature set in the training set into the pre-built multilayer feedforward neural network model to obtain the training prediction value;

[0141] Among them, the multilayer feedforward neural network (MLP) is a deep learning model that can extract complex nonlinear relationships of multidimensional features, and is suitable for risk prediction of latent faults in this scheme.

[0142] In some embodiments, the network structure of the multilayer feedforward neural network model includes an input layer, a hidden layer, and an output layer; wherein, the feature vector dimension of the input layer is consistent with the number of sample features; the hidden layer has two layers, with 64 and 32 nodes in each layer, and the activation function is ReLU; the output layer is a single node with the activation function Sigmoid, used to output the probability of latent fault risk, that is, the input features are transformed by linear transformation of weights and biases and nonlinear transformation of activation function to obtain the training prediction value.

[0143] Step S205: Compare the training prediction values ​​with the latent arc fault state labels in the training set, and calculate the cross-entropy loss function.

[0144] The cross-entropy loss function measures the difference between the predicted value and the true label, and is a commonly used loss function in classification tasks. The smaller the loss value, the closer the model's prediction is to the true label, providing feedback for updating the model parameters.

[0145] Step S206: Calculate the gradient and update the model parameters using the backpropagation algorithm until the cross-entropy loss function meets the preset requirements, thus obtaining the trained implicit arc fault prediction model.

[0146] In this algorithm, backpropagation calculates the gradient of the model parameters (weights and biases) based on the loss function, and then updates and adjusts the parameters using an optimizer (such as Adam). Through multiple iterations, the model is continuously optimized and eventually converges to the minimum value of the cross-entropy loss function, thus obtaining an accurate prediction model for hidden arc faults.

[0147] In the above implementation, the application of feature fusion and deep learning models can extract complex correlations between multi-dimensional features, output accurate fault risk probabilities, and dynamically update model parameters using cross-entropy loss and backpropagation algorithms to improve training efficiency. Ultimately, a highly efficient and accurate prediction model is obtained, providing decision support for intelligent operation and maintenance and resource optimization. This technical solution significantly improves the safety and intelligence level of 35KV collector lines in wind farms.

[0148] Reference Figure 3As a further implementation, after obtaining the trained implicit arc fault prediction model, the method further includes:

[0149] Step S301: Input the sample feature set in the test set into the trained latent arc fault prediction model to obtain the test prediction value;

[0150] After the model is trained, its generalization ability is verified through a test set to ensure that the model not only performs well on the training data, but also makes accurate predictions on unseen data. The sample feature set in the test set consists of data that was not used in the model optimization during training. After being input into the trained hidden arc fault prediction model, the model outputs a predicted value test for each sample, which is usually a fault risk probability value (e.g., a continuous value between 0 and 1).

[0151] Step S302: Compare the predicted test value with the latent arc fault status label in the test set to obtain the test comparison result;

[0152] Among these, test comparison results are the foundation for evaluating model performance. By comparing the model's predicted values ​​with the actual latent arc fault state labels in the test set, the model's predictive accuracy can be quantified. Specifically, test comparisons are typically measured using classification metrics (such as accuracy, precision, recall, and F1 score) and regression metrics (such as mean squared error).

[0153] Step S303: Based on the test comparison results, evaluate the performance index of the trained latent arc fault prediction model and iterate until the preset performance index is met or the preset number of iterations is reached, and obtain the trained latent arc fault prediction model.

[0154] Based on the test comparison results, the comprehensive performance index of the model is calculated and compared with the preset performance index. If the model performance does not meet the requirements, iterative optimization is carried out by adjusting the model structure, hyperparameters, etc.

[0155] In some embodiments, if the model performance does not meet the expected metrics, the model's hyperparameters (such as learning rate, regularization parameters, etc.) or structure (such as increasing the number of hidden layer nodes) are adjusted, and the model is retrained. The "training-testing-evaluation" process is repeated until the preset performance metrics are met or the maximum number of iterations is reached.

[0156] In the above implementation, when performance indicators fail to meet the standards, optimization is performed by adjusting the model structure and hyperparameters to ensure the reliability and generalization ability of the final model. An early stopping mechanism effectively avoids overfitting during model training, ensuring optimal performance on test data. Ultimately, the trained latent arc fault prediction model exhibits high accuracy and stability, enabling real-time and efficient early warning of latent faults in wind farm collector lines. This provides reliable guidance for the operation and maintenance team, improving the overall operational safety and efficiency of the wind farm.

[0157] Reference Figure 4 As a further implementation of the wind farm collector line fault early warning method, after the step of generating risk early warning information and risk maintenance strategies corresponding to dust accumulation risk sections based on the risk early warning level, it also includes:

[0158] Step S401: Prioritize the maintenance of dust-prone areas according to the risk warning level to obtain a maintenance task list including multiple priority groups;

[0159] Specifically, dust accumulation risk areas in wind farms are divided into different priority groups based on risk warning levels to ensure that high-risk areas are addressed first and optimize the scheduling of maintenance resources. The risk warning level is determined based on the fault risk probability output by the latent arc fault prediction model, and priority groups are formed through classification rules (such as high, medium, and low), with each priority group containing areas of the same risk level.

[0160] Step S402: Determine whether there are multiple dust accumulation risk sections in each priority group; if yes, proceed to step S403; if no, repeat step S402 for the next priority group.

[0161] Step S403: Based on the dust accumulation assessment index, the dust accumulation risk sections in the same priority group are sorted in a secondary order to obtain a secondary order list of the priority group.

[0162] If a priority group contains multiple sections, these sections need to be further refined into a secondary sort to optimize the maintenance sequence. The dust accumulation assessment index combines the current dust accumulation thickness and the dust accumulation growth rate, reflecting the severity and future trend of dust accumulation in a section. By using the dust accumulation assessment index to perform secondary sorting of multiple sections within the same priority group, sections with higher dust accumulation risk are prioritized for maintenance.

[0163] Understandably, multiple sections within the same priority group may have different dust accumulation states, thus requiring sorting based on more specific metrics (such as the dust accumulation assessment index). By ensuring that operations are performed only on priority groups that require further sorting, the use of computational resources is optimized, and redundant processing of single-section groups is avoided.

[0164] Step S404: Update the maintenance task list according to the secondary sorting list of each priority group;

[0165] In particular, by adding secondary sorting information to the maintenance task list, the segment order of the task list can simultaneously reflect the risk level and the priority of dust accumulation assessment, making the maintenance order more accurate and optimizing the execution efficiency of the maintenance plan.

[0166] Step S405: Based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk area, generate maintenance task information and send it to the maintenance terminal.

[0167] The updated maintenance task list is converted into specific maintenance task information and sent to the maintenance terminal for maintenance personnel to execute. The task information includes section ID, risk level, dust accumulation assessment index, maintenance content, priority, and other information to help the operations and maintenance team understand the task details.

[0168] The above implementation method achieves scientific maintenance and management of dust accumulation risk areas in wind farms. By using a dual ranking system based on risk warning levels and dust accumulation assessment indices, the priority and execution sequence of maintenance tasks are refined. Furthermore, dynamic maintenance strategies are combined to generate maintenance task information, providing precise guidance for operation and maintenance personnel. Through the optimization and real-time updating of the task list, not only is resource utilization efficiency improved, but the risk of latent fault propagation is also significantly reduced.

[0169] Reference Figure 5 As a further implementation of the wind farm collector line fault early warning method, after determining the risk warning level corresponding to the dust accumulation risk section based on the probability of latent arc fault risk, the method further includes:

[0170] Step S501: Determine whether the risk warning level corresponding to the dust accumulation risk section exceeds the preset level threshold; if yes, proceed to step S502; if no, do not perform any operation.

[0171] Step S502: Identify the dust accumulation risk area as a high-risk area;

[0172] Specifically, based on the risk warning level calculated from the probability of latent arc faults, high-risk sections that pose a significant threat to the safety of system operation are selected.

[0173] The preset risk level thresholds are set based on historical failure data and operational experience; for example, an 80% failure risk probability corresponds to a high-risk section. It should be noted that in special circumstances (such as extreme weather), the thresholds can be dynamically lowered to improve the sensitivity of risk response.

[0174] Step S503: Obtain the line topology of the wind farm collection line based on the preset section division information;

[0175] After identifying high-risk sections, their physical connections with other sections can be obtained through the topology, facilitating subsequent analysis of adjacent sections. The topology of the wind farm's collector lines reflects the physical relationships and electrical connection paths between each section.

[0176] Specifically, based on the GIS (Geographic Information System) data of the wind farm and the segmentation rules, a topology map of the line is constructed. The topology can be represented by an adjacency matrix, where the elements of the matrix reflect the connection relationship between segments. By traversing the adjacency matrix, the directly adjacent and indirectly adjacent segments of high-risk segments can be quickly located.

[0177] Step S504: Identify adjacent segments of high-risk segments based on the line topology and construct an adjacent segment list; wherein, the adjacent segment list includes directly adjacent segments and indirectly adjacent segments;

[0178] Directly adjacent sections refer to sections that are directly physically connected to high-risk sections, while indirectly adjacent sections refer to sections that have one or more levels of separation from high-risk sections. In other words, indirectly adjacent sections are not directly connected to high-risk sections, but are indirectly connected through one or more intermediate sections.

[0179] It should be noted that the number of interval levels between indirectly adjacent segments is limited (e.g., a maximum of two levels) to avoid excessive computational complexity due to an overly large analysis scope. By constructing a complete list of adjacent segments, a foundation is laid for subsequent feature similarity and synchronicity analysis.

[0180] Step S505: Based on the dust accumulation assessment index, calculate the similarity of dust accumulation characteristics between each adjacent section and the high-risk section.

[0181] Among them, the similarity between the dust accumulation assessment index of adjacent sections and high-risk sections is analyzed, and sections that are highly similar to high-risk sections are selected.

[0182] Specifically, the similarity formula can be used to calculate the similarity of dust accumulation characteristics between adjacent sections and high-risk sections:

[0183] .

[0184] Step S506: Based on a preset similarity threshold, determine highly similar segments of high-risk segments according to the similarity of dust accumulation features;

[0185] In this process, based on a pre-configured similarity threshold, adjacent sections that are similar to the dust accumulation characteristics of high-risk sections are selected, thus narrowing the analysis scope and providing support for subsequent fault propagation risk analysis.

[0186] Understandably, by calculating the similarity of dust accumulation characteristics, the analysis scope of high-risk sections is effectively extended to sections with highly similar dust accumulation characteristics, thus preserving sections that may have fault correlation. By introducing a similarity threshold mechanism, the complex problem of fault propagation path analysis is simplified into a similarity screening problem, significantly improving computational efficiency.

[0187] Step S507: Obtain historical electrical parameters of high-risk sections and perform preprocessing to obtain time series data of high-risk sections;

[0188] This involves extracting historical electrical parameter data (such as leakage current and electric field distortion) from high-risk sections and preprocessing the data to provide high-quality input for subsequent time series modeling. Abnormal fluctuations in electrical parameters are a significant indicator of latent arcing faults, and time series analysis can reveal these abnormal patterns.

[0189] Specifically, the data preprocessing steps include data cleaning and smoothing. This involves processing missing and outlier values ​​in the data, such as using interpolation to fill in missing data, denoising outliers, and then using a sliding window to smooth the data and eliminate the interference of random noise.

[0190] Step S508: Perform time series modeling based on time series data, conduct residual analysis and anomaly detection based on the ARIMA model, and determine the time points of abnormal fluctuations in high-risk areas;

[0191] The ARIMA model is suitable for time series with trends, seasonality, and random components, and can be used to capture these components and identify anomalous fluctuations. By using the ARIMA model to model the electrical parameter time series of high-risk areas, and then employing residual analysis, anomalous fluctuation points can be identified.

[0192] In one embodiment of this application, historical data can be used to train an ARIMA model to fit the time series of electrical parameters for high-risk sections:

[0193] ;

[0194] In the above formula, and These are the autoregressive and moving average coefficients, respectively.

[0195] Further, calculate the residuals. And determine whether it exceeds the preset abnormal threshold:

[0196] ;

[0197] like If so, it is marked as an abnormal fluctuation point.

[0198] It is understandable that time series modeling is used to capture abnormal fluctuations in electrical parameters in high-risk sections, and residual analysis is then used to automatically locate the time points of abnormal fluctuations in high-risk sections, providing a time reference for subsequent synchronization analysis.

[0199] Step S509: Obtain historical electrical parameters of highly similar sections at abnormal fluctuation time points and perform preprocessing;

[0200] Among these methods, electrical parameters of highly similar sections at abnormal fluctuation time points are extracted, and the extracted electrical parameter data is denoised and smoothed to ensure data quality. By introducing a time point matching mechanism in the fault propagation analysis, the analysis process becomes more logical and targeted.

[0201] Step S510: Based on the time points of abnormal fluctuations, perform electrical parameter synchronization analysis on high-risk sections and highly similar sections to obtain synchronization analysis results;

[0202] This involves analyzing whether the electrical parameters of high-risk sections and highly similar sections exhibit synchronous fluctuations at abnormal fluctuation points, thereby identifying potential fault propagation correlations. The results of the synchronization analysis and assessment are obtained by calculating the relative changes and correlations between electrical parameters.

[0203] Specifically, the synchronicity formula can be used to analyze the changes in electrical parameters (such as leakage current I) between the two sections:

[0204] ;

[0205] Specifically, by comparing the leakage current I of the two sections, correlation patterns are found, thereby accurately identifying the fault propagation correlation between high-risk sections and highly similar sections.

[0206] Step S511: Determine whether there is synchronous fluctuation of electrical parameters based on the synchronization analysis results. If yes, proceed to step S512; otherwise, do not perform any operation.

[0207] Step S512: Identify highly similar segments as highly correlated segments to obtain a list of highly correlated segments.

[0208] If the synchronicity analysis results exceed the preset threshold, it indicates that there is synchronous fluctuation, that is, the changes in electrical parameters of the two sections are highly correlated at the same point in time, thus identifying highly correlated sections that may have the risk of fault propagation.

[0209] Understandably, due to harsh environmental conditions (such as dust accumulation, strong winds, and humidity), latent arc faults in wind farm collector lines are a difficult-to-detect but potentially dangerous type of fault. Traditional methods typically analyze and handle faults only in single high-risk sections, neglecting the potential correlations between sections. This limitation leads to the failure to promptly detect and prevent potential risks of fault propagation. Therefore, this solution proposes a comprehensive approach based on the correlations between sections. It uses the similarity of dust accumulation characteristics to screen highly similar sections, analyzes the fluctuations of electrical parameters in high-risk sections based on time-series anomalies, and further combines synchronicity analysis to determine the fault propagation path. This aims to achieve accurate identification and efficient early warning of the risk of latent arc fault propagation in wind farm collector lines.

[0210] The above implementation achieves multi-level and multi-dimensional fault propagation risk analysis, enabling not only early warning of faults in single sections but also revealing potential risks of fault propagation, providing a scientific basis for intelligent maintenance of wind farm collection lines. Through real-time data-driven and intelligent analysis, this solution significantly improves the accuracy of fault management, reduces reliance on manual intervention, and promotes the intelligent and efficient operation and maintenance of wind farms.

[0211] Reference Figure 6 As one implementation of step S405, when the dust accumulation risk section is a high-risk section, the step of generating maintenance task information based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk section includes:

[0212] Step S601: Determine the list of corresponding highly correlated sections based on the dust accumulation risk sections;

[0213] Once a section is identified as a high-risk section, it is necessary to analyze the fault correlation between this section and other adjacent sections in order to identify highly correlated sections that have a potential risk of fault propagation. These highly correlated sections may have a potential risk of fault propagation with the high-risk section due to sharing similar environmental conditions or coupling of electrical parameters.

[0214] Step S602: Generate a synchronous maintenance task based on the list of highly correlated sections and add it to the maintenance strategy of the dust accumulation risk section to obtain the updated maintenance strategy.

[0215] Specifically, when maintaining high-risk sections, maintenance tasks for highly correlated sections are generated simultaneously to prevent fault propagation or hidden problems from being resolved in a timely manner. Since highly correlated sections are generally close to high-risk sections, they can be maintained together, thereby achieving efficient use of resources and reducing operation and maintenance costs.

[0216] It is understandable that highly correlated sections, due to their similarity or coupling with high-risk sections, may be affected simultaneously when problems occur in high-risk sections. Maintaining high-risk sections alone may not completely solve the problem, while maintaining highly correlated sections simultaneously can prevent potential failures from a global perspective.

[0217] Step S603: Based on the updated maintenance task list and the updated maintenance strategy, generate maintenance task information for dust accumulation risk areas.

[0218] After integrating the maintenance tasks for high-risk and highly correlated sections, complete maintenance task information is generated so that it can be sent to the maintenance terminal for efficient execution.

[0219] Specifically, highly correlated sections are considered more susceptible to failures in high-risk sections due to their significant similarities and synchronization with high-risk sections in terms of dust accumulation characteristics and abnormal electrical parameters. Therefore, their priority should be higher than ordinary sections but lower than directly high-risk sections. In terms of maintenance content, highly correlated sections should be subject to differentiated strategies based on their specific characteristics, such as prioritizing dust removal, electrical parameter testing, or inspection for localized hidden faults. Simultaneously, in conjunction with the maintenance tasks of high-risk sections, synchronous inspections and preventative maintenance can be performed on highly correlated sections to effectively block potential fault propagation paths, thereby achieving a dual optimization of resource utilization efficiency and fault prevention effectiveness.

[0220] The above implementation successfully addresses the lack of a holistic perspective in traditional wind farm maintenance. By extending maintenance of high-risk sections to related sections, it systematically prevents the propagation of potential faults, significantly improving the operation and maintenance efficiency and safety of wind farms. The complete closed loop from fault identification to global maintenance effectively enhances the reliability and intelligent operation and maintenance level of wind farm power collection lines.

[0221] This application also discloses a fault early warning system for wind farm collection lines.

[0222] A fault early warning system for wind farm collector lines, the fault early warning system comprising:

[0223] The segment division module is used to divide the wind farm's collection lines into multiple segments based on preset segment division information.

[0224] The monitoring module is used to acquire the current monitoring data of each section in real time; the current monitoring data includes the current electrical parameters, current environmental data, and current dust accumulation thickness data;

[0225] The judgment module is used to determine whether there is a fault in each section based on the current electrical parameters; if yes, it outputs the first judgment result; if no, it outputs the second judgment result.

[0226] The fault indication module is used to determine the section as a faulty section in response to the first judgment result and send fault maintenance prompt information to the management terminal.

[0227] The dynamic deviation calculation module is used to calculate the dynamic deviation between the current environmental data of the section and the environmental baseline data to obtain the dynamic environmental feature vector of the section.

[0228] The dust accumulation growth prediction module is used to calculate the predicted dust accumulation growth rate of a section based on a preset algorithm, by combining dynamic environmental feature vectors.

[0229] The dust accumulation assessment module is used to calculate the dust accumulation assessment index of a section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate.

[0230] The dust accumulation risk zone determination module is used to determine the dust accumulation risk zone as the dust accumulation assessment index exceeding the preset threshold.

[0231] The feature vector fusion module is used to fuse the current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vectors of the dust accumulation risk zone to generate a comprehensive fault feature vector.

[0232] The latent arc fault prediction module is used to input the comprehensive fault feature vector of the dust accumulation risk section into the pre-built latent arc fault prediction model to obtain the latent arc fault risk probability of the dust accumulation risk section.

[0233] The warning level determination module is used to determine the risk warning level corresponding to the dust accumulation risk section based on the probability of hidden arc fault risk.

[0234] The risk warning and maintenance module is used to generate risk warning information and risk maintenance strategies for dust accumulation risk sections based on the risk warning level.

[0235] In the above implementation, multi-dimensional monitoring data, including environmental, dust accumulation, and electrical parameters, are combined to form a complete early warning chain from fault diagnosis to latent arc fault risk prediction. The system can not only monitor and analyze the status of wind farm collector lines in real time, but also accurately assess dust accumulation risk and its potential impact on electrical performance through innovative technologies such as dynamic deviation calculation, dust growth prediction, and feature vector fusion. Based on the risk probability analysis of the latent arc fault prediction model, the system further determines the risk warning level and generates corresponding maintenance strategies, realizing a transformation from single fault detection to comprehensive risk assessment and maintenance optimization.

[0236] The wind farm collector line fault early warning system of this application embodiment can implement any of the above-mentioned fault early warning methods, and the specific working process of each module in the fault early warning system can refer to the corresponding process in the above-mentioned method embodiments.

[0237] In the several embodiments provided in this application, it should be understood that the provided methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for example, the division of a certain module is merely a logical functional division, and in actual implementation there may be other division methods, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed.

[0238] This application also discloses a computer device.

[0239] A computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement a wind farm collector line fault early warning method as described above.

[0240] This application also discloses a computer-readable storage medium.

[0241] A computer-readable storage medium storing a computer program that can be loaded by a processor and executed as described above in any of the wind farm collector line fault early warning methods.

[0242] The computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device; the program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0243] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0244] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for early warning of faults in wind farm collector lines, characterized in that, The method includes: The wind farm's power collection line is divided into multiple sections according to the preset section division information; Real-time acquisition of current monitoring data for each section; the current monitoring data includes current electrical parameters, current environmental data, and current dust accumulation thickness data; Based on the current electrical parameters, determine whether there is a fault in each section; if so, identify the section as a faulty section and send a fault maintenance prompt message to the management terminal. If not, the dynamic deviation between the current environmental data of the segment and the environmental baseline data is calculated to obtain the dynamic environmental feature vector of the segment. Based on the dynamic environmental feature vector, the predicted dust accumulation growth rate of the section is calculated using a preset algorithm. Calculate the dust accumulation assessment index of the section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate of the section; The section where the dust accumulation assessment index exceeds a preset threshold is identified as a dust accumulation risk section; The current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vector of the dust accumulation risk zone are fused to generate a comprehensive fault feature vector. The comprehensive fault feature vector of the dust accumulation risk section is input into the pre-constructed hidden arc fault prediction model to obtain the hidden arc fault risk probability of the dust accumulation risk section. Based on the probability of hidden arc fault risk, determine the risk warning level corresponding to the dust accumulation risk section; Based on the risk warning level, generate risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section.

2. The method for early warning of faults in wind farm collector lines according to claim 1, characterized in that, The formula for calculating the predicted dust accumulation growth rate of the aforementioned section is as follows: ， In the above formula, ΔV, ΔH, and ΔT represent wind speed deviation, humidity deviation, and temperature deviation, respectively. 基线 H 基线 T 基线 These are the wind speed baseline, humidity baseline, and temperature baseline, respectively. 当前湿度 The current humidity data is given, and γ1, γ2, γ3, and γ4 are empirical adjustment coefficients, respectively.

3. The method for early warning of faults in wind farm collector lines according to claim 1, characterized in that, It also includes a training step for the latent arc fault prediction model, the training step comprising: Acquire historical datasets and perform preprocessing; the historical datasets include historical monitoring data and pre-labeled latent arc fault status tags, and the historical monitoring data includes historical dust accumulation thickness data, historical electrical parameters, historical dust accumulation growth rate and historical environmental data; The historical monitoring data are fused with feature vectors to generate a sample feature set, which is then associated with the latent arc fault status label. The sample feature set after label association is divided into a training set and a test set; The sample feature set in the training set is input into a pre-constructed multi-layer feedforward neural network model to obtain training prediction values; The training prediction values ​​are compared with the latent arc fault state labels in the training set, and the cross-entropy loss function is calculated. The gradient is calculated and the model parameters are updated by backpropagation until the cross-entropy loss function meets the preset requirements, thus obtaining the trained implicit arc fault prediction model.

4. The method for early warning of faults in wind farm collector lines according to claim 3, characterized in that, The step following obtaining the trained implicit arc fault prediction model also includes: The sample feature set in the test set is input into the trained latent arc fault prediction model to obtain the test prediction value; The test prediction value is compared with the latent arc fault status label in the test set to obtain the test comparison result; Based on the test comparison results, the performance index of the trained latent arc fault prediction model is evaluated and iterated until the preset performance index is met or the preset number of iterations is reached, thus obtaining the trained latent arc fault prediction model.

5. A method for early warning of faults in wind farm collector lines according to any one of claims 1 to 4, characterized in that, After the step of generating risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section based on the risk warning level, the method further includes: Based on the risk warning level, the maintenance priority of the dust accumulation risk area is sorted to obtain a maintenance task list including multiple priority groups; Determine whether there are multiple dust accumulation risk zones in each priority group; If so, the dust risk zones in the same priority group are then sorted in a secondary order according to the dust accumulation assessment index to obtain a secondary order list of the priority group. The maintenance task list is updated according to the secondary sorting list of each priority group; Based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk area, maintenance task information is generated and sent to the maintenance terminal.

6. A method for early warning of faults in wind farm collector lines according to claim 5, characterized in that, After determining the risk warning level corresponding to the dust accumulation risk section based on the probability of the latent arc fault risk, the method further includes: Determine whether the risk warning level corresponding to the dust accumulation risk section exceeds the preset level threshold. If so, the dust accumulation risk section is determined as a high-risk section. The topology of the wind farm's power collection lines is obtained based on the preset segment division information; Based on the line topology, adjacent segments of the high-risk segment are identified, and an adjacent segment list is constructed; the adjacent segment list includes directly adjacent segments and indirectly adjacent segments; Based on the dust accumulation assessment index, the similarity of dust accumulation characteristics between each adjacent segment and the high-risk segment is calculated. Based on a preset similarity threshold, highly similar segments of the high-risk segments are determined according to the similarity of the dust accumulation features; Historical electrical parameters of the high-risk section are obtained and preprocessed to obtain time series data of the high-risk section; Time series modeling is performed based on the time series data, and residual analysis and anomaly detection are conducted based on the ARIMA model to determine the time points of abnormal fluctuations in the high-risk segment. Obtain the historical electrical parameters of the highly similar section at the time point of the abnormal fluctuation and perform preprocessing; Based on the abnormal fluctuation time points, electrical parameter synchronization analysis is performed on the high-risk section and the highly similar section to obtain the synchronization analysis results; Based on the results of the synchronization analysis, it is determined whether there is synchronous fluctuation of electrical parameters. If so, the highly similar segments are identified as highly correlated segments, and a list of highly correlated segments is obtained.

7. A method for early warning of faults in wind farm collector lines according to claim 6, characterized in that, When the dust accumulation risk zone is a high-risk zone, the steps for generating maintenance task information based on the updated maintenance task list and the maintenance strategy corresponding to the dust accumulation risk zone include: Determine the corresponding list of highly correlated zones based on the dust accumulation risk zones; Synchronous maintenance tasks are generated based on the list of highly correlated sections and added to the maintenance strategy of the dust accumulation risk section to obtain the updated maintenance strategy. Based on the updated maintenance task list and the updated maintenance strategy, maintenance task information for the dust accumulation risk area is generated.

8. A fault early warning system for wind farm collector lines, characterized in that, The system includes: The segment division module is used to divide the wind farm collection line into multiple segments according to preset segment division information; The monitoring module is used to acquire the current monitoring data of each section in real time; the current monitoring data includes current electrical parameters, current environmental data, and current dust accumulation thickness data. The judgment module is used to determine whether there is a fault in each section based on the current electrical parameters; if so, it outputs a first judgment result; if not, it outputs a second judgment result. The fault indication module is used to determine the segment as a faulty segment in response to the first judgment result and send fault maintenance prompt information to the management terminal. The dynamic deviation calculation module is used to perform dynamic deviation calculation between the current environmental data of the section and the environmental baseline data to obtain the dynamic environmental feature vector of the section. The dust growth prediction module is used to combine the dynamic environmental feature vector and calculate the predicted dust growth rate of the section based on a preset algorithm. The dust accumulation assessment module is used to calculate the dust accumulation assessment index of the section based on the current dust accumulation thickness data and the predicted dust accumulation growth rate of the section. The dust accumulation risk zone determination module is used to determine the zone where the dust accumulation assessment index exceeds a preset threshold as a dust accumulation risk zone. The feature vector fusion module is used to fuse the current dust thickness data, current electrical parameters, predicted dust growth rate, and dynamic environmental feature vectors of the dust accumulation risk section to generate a comprehensive fault feature vector. The latent arc fault prediction module is used to input the comprehensive fault feature vector of the dust accumulation risk section into the pre-constructed latent arc fault prediction model to obtain the latent arc fault risk probability of the dust accumulation risk section. The early warning level determination module is used to determine the risk early warning level corresponding to the dust accumulation risk section based on the probability of the hidden arc fault risk. The risk warning and maintenance module is used to generate risk warning information and risk maintenance strategies corresponding to the dust accumulation risk section based on the risk warning level.

9. A computer device, characterized in that: The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer program is stored that can be loaded by a processor and executed as described in any one of claims 1 to 7.

Images