A 5G offshore wind farm intelligent inspection regulation method, system, device and storage medium

Abstract

The application provides a 5G offshore wind farm intelligent inspection regulation method, system, device and storage medium, which comprises the following steps: S1, collecting real-time fault prediction data of the wind farm through a multi-source heterogeneous sensor deployed in the offshore wind farm; S2, transmitting the data to an edge computing node in real time through a 5G communication network to complete data preprocessing; S3, uploading the preprocessed real-time prediction data to a cloud analysis platform, extracting features by using a multi-modal LSTM, and generating a fault risk membership portrait of the wind turbine based on a fuzzy C-means clustering algorithm; S4, constructing a multi-objective optimization function according to the risk membership, and generating a globally optimal unmanned inspection path by using a genetic algorithm. The application realizes efficient data perception through multi-source sensors and edge computing, significantly improves the inspection efficiency and fault recognition accuracy by using soft partition risk assessment and global heuristic optimization, and has self-adaptive learning ability to support long-term stable operation.

Description

Technical Field

[0001] This invention belongs to the field of communication technology and remote operation and maintenance of wind power equipment, specifically involving a 5G offshore wind farm intelligent inspection and control method, system, equipment and storage medium. Background Technology

[0002] Offshore wind power, as one of the key clean energy forms currently being developed globally, is being rapidly deployed on a large scale in my country. As of the end of April 2025, my country's cumulative installed capacity of offshore wind power had reached 43.51 million kilowatts, maintaining its leading position globally. With the continuous expansion of the number and scale of offshore wind farms, the demand for remote operation and maintenance and intelligent management of wind power equipment has increased significantly, and inspection systems are facing higher requirements for accuracy, responsiveness, and safety.

[0003] Traditional offshore wind power operation and maintenance methods largely rely on manual tower climbing and support vessels, resulting in long operation cycles, low efficiency, high safety risks, and heavy costs. In recent years, with the development of key technologies such as 5G communication, differential positioning, and edge computing, various unmanned inspection methods have been gradually applied to the offshore wind power field. These include drones, automated robots, and remote sensing platforms, which offer advantages such as remote control, real-time information transmission, and flexible deployment, gradually replacing traditional manual inspection methods.

[0004] However, existing unmanned inspection systems typically employ a unified route planning approach for the entire wind farm, failing to adequately consider the differences in operating status and maintenance cycles among wind turbines. Some newly built turbines or those recently overhauled are in a low-risk state in the short term, and their inspection frequency can be appropriately reduced, while older turbines or those with abnormal operating parameters should have their inspection frequency increased. Unified route scheduling may not only waste inspection resources but also delay the discovery and handling of high-risk equipment, impacting overall operational efficiency and safety.

[0005] Furthermore, offshore wind turbines operate in environments constantly exposed to high humidity, salt spray, sandstorms, and ultraviolet radiation. The blades, in particular, are prone to structural damage such as scratches, cracks, and material aging, which can even lead to shutdowns in severe cases. Extreme weather events like typhoons and strong waves can also cause additional stress on the turbines, increasing the volatility and uncertainty of their operation. Currently, most inspection systems do not adequately correlate environmental changes with the turbine's health status, lacking an effective mechanism to perceive the continuous impact of external sea conditions on the operational risks of wind turbines. Summary of the Invention

[0006] The first objective of this invention is to provide a 5G-based intelligent inspection and control method for offshore wind farms, addressing the aforementioned problems.

[0007] Therefore, the above-mentioned objective of the present invention is achieved through the following technical solution:

[0008] A method for intelligent inspection and control of 5G offshore wind farms includes the following steps:

[0009] S1. Collect real-time fault prediction data of wind farms, including wind turbine status parameters and environmental parameters, through multi-source heterogeneous sensors deployed at offshore wind farms.

[0010] S2. Based on the 5G communication network, the fault prediction data is transmitted to the edge computing node in real time, and the edge computing node performs preprocessing operations on the fault prediction data.

[0011] S3. Upload the preprocessed real-time prediction data to the cloud analysis platform, and use a multimodal long short-term memory network fault prediction model with an attention mechanism to extract key time series features and generate a fault risk membership profile for each wind turbine based on a fuzzy clustering algorithm.

[0012] S4. Based on the risk membership degree output by the cloud model, construct a multi-objective optimization function and use a genetic algorithm to generate the globally optimal unmanned inspection path and scheduling strategy.

[0013] S5. Based on historical operating conditions and fault feedback data during the actual operation of the wind turbine, the fault prediction model is updated offline on a regular basis.

[0014] While adopting the above technical solutions, the present invention may also adopt or combine the following technical solutions:

[0015] As a preferred technical solution of the present invention: in step S1, the fan status parameters are obtained through vibration sensor, temperature sensor, blade speed sensor and sound collector.

[0016] As a preferred technical solution of the present invention: in step S1, environmental parameters are obtained by an anemometer, a temperature and humidity sensor, and an aerosol lidar.

[0017] As a preferred technical solution of the present invention, step S2 specifically includes the following steps:

[0018] S21. Perform data cleaning operations on the operating status parameters and environmental information of each wind turbine, including outlier detection, missing value completion and format standardization.

[0019] S22. The original sequence is reconstructed by using a sliding time window mechanism, dividing it into fixed-length time segments for use as input for time series modeling and prediction in the cloud.

[0020] As a preferred technical solution of the present invention, step S3 specifically includes the following steps:

[0021] S31. Input Construction: After the sliding window reconstruction of the wind turbine status and environmental data is completed at the edge computing node, the obtained input samples are organized into two synchronously aligned time series matrices:

[0022] Wind turbine state parameter matrix, Where T represents the length of the time window, These include: fan vibration frequency, gearbox temperature, fan blade speed, and sound decibels;

[0023] Environmental parameter matrix Where T represents the length of the time window, This includes: wind speed, air temperature, air humidity, and sea salt particle concentration;

[0024] S32 and LSTM timing modeling were performed, using two LSTM networks respectively. , After processing, we get:

[0025] Hidden sequence of state parameters: ,

[0026] Hidden sequence of environmental parameters: ;

[0027] S33. Attention score calculation: For each time step of the hidden state in the hidden sequence of each of the two channels... Calculate attention score :

[0028]

[0029] In the formula, , , These are trainable parameter vectors and weight matrices;

[0030] S34. Attention weight normalization and context aggregation: The score vector is normalized to obtain the weight coefficients. :

[0031]

[0032] This is used to perform weighted aggregation of the sequences and calculate the channel context feature vector:

[0033]

[0034] The state parameter channel context vectors are obtained respectively. With environmental parameter channel context vector ;

[0035] S35. Feature fusion and risk prediction output: The two attention context vectors are concatenated to form a feature fusion vector. The input is fed into the fully connected layer to generate a fault probability prediction value. And normalized using the Sigmoid activation function;

[0036] S36. Construction of fuzzy risk profiles, using the fuzzy C-means clustering algorithm (FCM) to assess failure probabilities. And process the feature fusion vector:

[0037] Set cluster center set These correspond to the normal state, the potential hidden danger state, and the serious fault state, respectively.

[0038] Calculate the membership degree of each wind turbine node i to risk category j. This forms the risk membership matrix U, which serves as the input for downstream regulatory weights.

[0039]

[0040] In the formula, Let be the feature vector of wind turbine i, and m be the fuzzy weighted index.

[0041] As a preferred technical solution of the present invention, step S32 specifically includes:

[0042] S321. Set the hyperparameter set:

[0043] In the formula, These represent the number of LSTM cells in the wind turbine state network and the offshore environment network, respectively. For learning rate, The L2 regularization coefficient is... Batch size;

[0044] S322. Define the binary cross-entropy loss function on the validation set as the optimization objective;

[0045] S323. Establish hyperparameters using a Gaussian process surrogate model. To determine the optimal hyperparameter combination based on the relationship between the validation loss and the expected improvement (EI) or probabilistic improvement (PI) strategies, the following approach can be used:

[0046] .

[0047] As a preferred technical solution of the present invention, step S4 specifically includes the following steps:

[0048] S41. Problem coding: Constructing chromosomes for the genetic algorithm using integer coding, chromosome sequence. The order in which the unmanned inspection equipment visits the fans is represented. Number the wind turbine;

[0049] S42. Define a fitness function F with the goal of maximizing risk capture benefits and minimizing inspection costs:

[0050]

[0051] In the formula, These are the weighting coefficients. Let the membership degree of fan k belong to the severe fault class be denoted as . The time decay factor, The time to access wind turbine k, The Euclidean distance between the two wind turbines;

[0052] S43. Genetic evolution: Perform selection, crossover, and mutation operations on the initial population.

[0053] Selection operation: Use the roulette wheel algorithm to select individuals with high fitness;

[0054] Crossover operation: The sequential crossover method OX is used to generate the child path while preserving the risk priority sequence characteristics of the parent generation;

[0055] Mutation operation: Randomly swap the positions of two genes in a chromosome sequence to escape local optima;

[0056] S44. Strategy Output: When the number of iterations reaches a preset value, output the chromosome sequence with the highest fitness as the final unmanned inspection path.

[0057] The second objective of this invention is to provide a 5G intelligent inspection and control system for offshore wind farms, comprising the following modules:

[0058] The sensing and acquisition module is used to collect wind power equipment operating status parameters and environmental information in real time through multi-source heterogeneous sensors deployed on the offshore wind turbine and its operating environment.

[0059] The network transmission module is used to transmit the sensed data to the edge computing node at high speed and stably through the 5G communication module, and supports edge access gateway, clock synchronization and link QoS guarantee functions;

[0060] The edge preprocessing module is used to clean, remove outliers, fill in missing data, convert formats, and reconstruct sliding time windows for received data at edge computing nodes, and then upload the structured time series data to the cloud platform.

[0061] The cloud-based analytics module is used to perform multimodal LSTM modeling and attention mechanism fusion, and includes fuzzy clustering units to generate membership matrices of wind turbines for different risk levels.

[0062] The scheduling and control module integrates a genetic algorithm optimization engine, which is used to calculate the globally optimal inspection path and issue control commands based on the risk membership matrix and the geographical distribution of the wind field.

[0063] The application business module is used to display a visual map of wind farm fault risks, configure remote operation and maintenance tasks, and provide operation and maintenance support functions such as remote feedback monitoring and historical record management.

[0064] A third objective of this invention is to provide an electronic device comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus.

[0065] The memory is used to store computer programs;

[0066] A processor is used to execute a computer program stored in a memory to implement the steps of the 5G offshore wind farm intelligent inspection and control method described above.

[0067] Another objective of this invention is to provide a non-volatile storage medium storing an executable program, which, when executed by a processor, implements the steps of the 5G offshore wind farm intelligent inspection and control method described above.

[0068] Compared with the prior art, the present invention has the following beneficial effects:

[0069] 1) This invention addresses the problems of ambiguous fault definition and limited inspection resources in offshore wind farms by proposing an intelligent control method that integrates multimodal temporal modeling, fuzzy clustering risk assessment, and global scheduling using a genetic algorithm. Unlike traditional technologies, this invention achieves a "soft assessment" of wind turbine health status through risk membership and utilizes a multi-objective genetic algorithm to solve for the optimal inspection path, thereby maximizing the coverage efficiency of a single inspection operation while ensuring that high-risk equipment is prioritized for handling.

[0070] 2) This invention achieves synchronous perception of the wind turbine's operating status and external environment by deploying multi-source heterogeneous sensors, and completes near-source preprocessing and efficient uploading of data based on a collaborative architecture of 5G communication and edge computing. This reduces network and cloud load while ensuring timeliness, and improves the overall system's data processing capability and real-time response level.

[0071] 3) This invention innovatively introduces the Fuzzy C-means Clustering (FCM) algorithm to replace the traditional hard threshold judgment. Existing technologies typically set fixed thresholds (e.g., 0.8), which mathematically introduce truncation errors and easily lead to wind turbines in a critical state (e.g., a failure probability of 0.79) being misclassified as low-risk. This invention establishes a continuous risk assessment system by calculating the membership degree of wind turbines to each risk center, fundamentally eliminating the classification ambiguity of critical values. Compared to traditional methods, this method is expected to reduce the missed detection rate of early minor faults by approximately 10% to 15%, thereby achieving earlier early warning during the golden window period before a fault occurs.

[0072] 4) This invention establishes a global dynamic inspection and control strategy based on a genetic algorithm (GA), constructing a multi-objective fitness function that includes "risk capture reward" and "path movement cost". Compared to the rigid "hierarchical round-robin" or simple "greedy rule" (i.e., only considering the nearest high-risk point) in existing technologies, this strategy can search for the globally optimal solution in a vast solution space. Under the same inspection resource constraints, this method is expected to improve the average response speed of high-risk equipment by about 20%~30% and reduce the invalid inspection range and energy consumption by about 15%, achieving a significant optimization of inspection efficiency.

[0073] 5) This invention can periodically update the model by combining historical operating data, has adaptive learning capabilities, and can continuously adapt to different wind farm scenarios and wind turbine aging characteristics, thereby achieving long-term stable operation and enhanced risk identification capabilities, providing a solid data foundation and intelligent support for the intelligent and unmanned operation and maintenance of offshore wind farms. Attached Figure Description

[0074] Figure 1 The flowchart shows the intelligent inspection and control method for 5G offshore wind farms provided by this invention.

[0075] Figure 2 This is a detailed structure and calculation flowchart of a multimodal fault prediction model.

[0076] Figure 3 This is a system architecture diagram. Detailed Implementation

[0077] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0078] like Figure 1 The diagram illustrates the complete workflow from multi-source data acquisition, edge preprocessing, cloud prediction to control strategy output. A smart inspection and control method for 5G offshore wind farms specifically includes the following steps:

[0079] S1. Collect real-time fault prediction data of wind farms, including wind turbine status parameters and environmental parameters, through multi-source heterogeneous sensors deployed at offshore wind farms.

[0080] Wind turbine status parameters are obtained through vibration sensors, temperature sensors, blade speed sensors, and sound collectors installed on each wind turbine, while environmental parameters are obtained through anemometers, temperature and humidity sensors, and aerosol lidar deployed around the offshore wind farm and at the tower base.

[0081] Vibration sensors are used to monitor the mechanical vibration characteristics of wind turbines during operation.

[0082] Temperature sensors are used to collect temperature changes in key components of wind turbines or the environment inside the nacelle.

[0083] Blade speed sensor, used to obtain the rotational speed of wind turbine blades;

[0084] A sound acquisition device is used to collect acoustic signals during the operation of a wind turbine.

[0085] An anemometer is used to measure wind speed at sea.

[0086] Temperature and humidity sensors are used to collect air temperature and relative humidity.

[0087] Aerosol lidar is used to monitor the concentration of suspended particulate matter in the air, specifically the concentration of aerosol particles in the air, and is particularly used to detect the content of sea salt particles to assess the intensity of salt spray corrosion.

[0088] The aforementioned sensors perform preliminary integration of the collected signals through a local embedded control unit, forming a set of wind turbine status parameters and a set of marine environmental parameters, which are then uploaded to the edge computing node via a 5G communication module.

[0089] S2. Based on the 5G communication network, the fault prediction data is transmitted to the edge computing node in real time, and the edge computing node performs preprocessing operations on the fault prediction data.

[0090] S21. Perform data cleaning operations on the operating status parameters and environmental information of each wind turbine, including outlier detection, missing value completion and format standardization.

[0091] S22. The original sequence is reconstructed by using a sliding time window mechanism, dividing it into fixed-length time segments for use as input for time series modeling and prediction in the cloud.

[0092] S3. Upload the preprocessed real-time prediction data to the cloud analysis platform. Use a multimodal long short-term memory (LSTM) fault prediction model incorporating an attention mechanism to extract key temporal features and create fault risk membership profiles for each wind turbine. Figure 2The diagram illustrates the operational logic of key model components such as the dual-channel LSTM sub-network for wind turbine status and environmental parameters, the attention mechanism module, context feature fusion, and risk probability output.

[0093] S31. Input Construction: After the sliding window reconstruction of the wind turbine status and environmental data is completed at the edge computing node, the obtained input samples are organized into two synchronously aligned time series matrices:

[0094] Wind turbine state parameter matrix, Where T represents the length of the time window, These include: fan vibration frequency, gearbox temperature, fan blade speed, and sound decibels;

[0095] Environmental parameter matrix Where T represents the length of the time window, This includes: wind speed, air temperature, air humidity, and sea salt particle concentration;

[0096] S32 and LSTM timing modeling were performed, using two LSTM networks respectively. , After processing, we get:

[0097] Hidden sequence of state parameters: ,

[0098] Hidden sequence of environmental parameters: ;

[0099] S321. Set the hyperparameter set:

[0100] In the formula, These represent the number of LSTM cells in the wind turbine state network and the offshore environment network, respectively. For learning rate, The L2 regularization coefficient is... Batch size;

[0101] S322. Define the binary cross-entropy loss function on the validation set as the optimization objective;

[0102] S323. Establish hyperparameters using a Gaussian process surrogate model. To determine the optimal hyperparameter combination based on the relationship between the validation loss and the expected improvement (EI) or probabilistic improvement (PI) strategies, the following approach can be used:

[0103] .

[0104] S33. Attention score calculation: For each time step of the hidden state in the hidden sequence of each of the two channels... Calculate attention score :

[0105]

[0106] In the formula, , , These are trainable parameter vectors and weight matrices;

[0107] S34. Attention weight normalization and context aggregation: The score vector is normalized to obtain the weight coefficients. :

[0108]

[0109] This is used to perform weighted aggregation of the sequences and calculate the channel context feature vector:

[0110]

[0111] The state parameter channel context vectors are obtained respectively. With environmental parameter channel context vector ;

[0112] S35. Feature fusion and risk prediction output: The two attention context vectors are concatenated to form a feature fusion vector. The input is fed into the fully connected layer to generate a fault probability prediction value. And normalized using the Sigmoid activation function;

[0113] S36. Construction of fuzzy risk profiles, using the fuzzy C-means clustering algorithm (FCM) to assess failure probabilities. And process the feature fusion vector:

[0114] Set cluster center set These correspond to the normal state, the potential hidden danger state, and the serious fault state, respectively.

[0115] Calculate the membership degree of each wind turbine node i to risk category j. This forms the risk membership matrix U, which serves as the input for downstream regulatory weights.

[0116]

[0117] In the formula, Let be the feature vector of wind turbine i, and m be the fuzzy weighted index.

[0118] S4. Based on the risk membership degree output by the cloud model, construct a multi-objective optimization function and use a genetic algorithm to generate the globally optimal unmanned inspection path and scheduling strategy.

[0119] S41. Problem coding: Constructing chromosomes for the genetic algorithm using integer coding, chromosome sequence. The order in which the unmanned inspection equipment visits the fans is represented. Number the wind turbine;

[0120] S42. Define a fitness function F with the goal of maximizing risk capture benefits and minimizing inspection costs:

[0121]

[0122] In the formula, These are the weighting coefficients. Let the membership degree of fan k belong to the severe fault class be denoted as . The time decay factor, The time to access wind turbine k, The Euclidean distance between the two wind turbines;

[0123] S43. Genetic evolution: Perform selection, crossover, and mutation operations on the initial population.

[0124] Selection operation: Use the roulette wheel algorithm to select individuals with high fitness;

[0125] Crossover operation: The sequential crossover method OX is used to generate the child path while preserving the risk priority sequence characteristics of the parent generation;

[0126] Mutation operation: Randomly swap the positions of two genes in a chromosome sequence to escape local optima;

[0127] S44. Strategy Output: When the number of iterations reaches a preset value, output the chromosome sequence with the highest fitness as the final unmanned inspection path.

[0128] S5. Based on historical operating conditions and fault feedback data during the actual operation of the wind turbine, the fault prediction model is updated offline regularly to maintain the model's prediction accuracy and generalization ability.

[0129] The model update process includes extracting new sample data from the operation and maintenance platform, retraining the model parameters, and deploying the updated model version to improve the ability to identify risks in the wind turbine operating status.

[0130] This periodic update mechanism ensures that the model maintains good predictive performance and generalization ability despite changes in the wind farm operating environment, equipment aging, or maintenance practices.

[0131] To preliminarily verify the feasibility and effectiveness of the intelligent inspection and control method for 5G offshore wind farms proposed in this invention, a numerical simulation environment containing 30 wind turbines with a distribution range of 5 km × 5 km was constructed, and the "risk identification capability" and "inspection scheduling efficiency" were theoretically calculated based on typical working conditions.

[0132] 1) Measurement of the improvement in risk identification capability

[0133] In terms of fault risk identification, the simulation model first simulates the feature fusion process of multimodal sensor data. The results show that the attention mechanism proposed in this invention can effectively achieve feature focusing. When processing specific fault samples, it can automatically assign higher weights to key abnormal parameters such as vibration and temperature, thereby providing accurate input for subsequent risk assessment. Furthermore, for critical samples in the early stage of fault evolution (fault probability of approximately 0.79), the hard threshold judgment method in the existing technology fails due to the truncation effect, resulting in a high false negative rate. However, the FCM fuzzy clustering algorithm used in this invention successfully achieves soft partitioning and identification of potential hazards by calculating the membership degree of samples to the center of severe faults. Simulation verification shows that when processing such edge samples, the method of this invention increases the fault recall rate from less than 42.9% of the traditional method to 100%, achieving "zero false negatives" for high-risk hazards and greatly improving the safety early warning capability of offshore wind farms.

[0134] 2) Optimization calculation of inspection and dispatch efficiency

[0135] In the scheduling of UAV inspections, traditional scheduling strategies often focus solely on spatial distance, neglecting the urgency and priority of equipment failure risks. This easily leads to high-risk targets being overlooked due to their distance. This invention utilizes a genetic algorithm for global optimization, constructing a multi-objective fitness function that incorporates risk priority weights. In scheduling decisions, it mandates priority response to faulty units with high membership. Simulation data shows that, under the same range constraint, compared to the short-sighted path of traditional strategies that gets trapped in local optima, the globally optimal path generated by this method can traverse low-value areas and directly reach high-risk nodes. Measurement results demonstrate that this method significantly improves risk capture efficiency by approximately 85.9% per unit range, effectively avoiding resource waste caused by inefficient inspections and achieving precise and efficient allocation of maintenance resources to high-risk equipment.

[0136] This invention also provides a 5G offshore wind farm intelligent inspection and control system, such as... Figure 3 The diagram illustrates the system's layered structure and functional module division, from the perception layer, network transmission layer, edge computing layer, cloud computing layer, control and regulation layer to the application service layer. The modules include the following:

[0137] The sensing and acquisition module is used to collect wind power equipment operating status parameters and environmental information in real time through multi-source heterogeneous sensors deployed on the offshore wind turbine and its operating environment.

[0138] The network transmission module is used to transmit the sensed data to the edge computing node at high speed and stably through the 5G communication module, and supports edge access gateway, clock synchronization and link QoS guarantee functions;

[0139] The edge preprocessing module is used to clean, remove outliers, fill in missing data, convert formats, and reconstruct sliding time windows for received data at edge computing nodes, and then upload the structured time series data to the cloud platform.

[0140] The cloud-based analytics module is used to perform multimodal LSTM modeling and attention mechanism fusion, and includes fuzzy clustering units to generate membership matrices of wind turbines for different risk levels.

[0141] The scheduling and control module integrates a genetic algorithm optimization engine, which is used to calculate the globally optimal inspection path and issue control commands based on the risk membership matrix and the geographical distribution of the wind field.

[0142] The application business module is used to display a visual map of wind farm fault risks, configure remote operation and maintenance tasks, and provide operation and maintenance support functions such as remote feedback monitoring and historical record management.

[0143] The present invention also provides an electronic device, which includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other via the communication bus.

[0144] Memory, used to store computer programs;

[0145] The processor is used to execute computer programs stored in memory to implement the steps of the 5G offshore wind farm intelligent inspection and control method described above.

[0146] The present invention also provides a non-volatile storage medium storing an executable program, which, when executed by a processor, implements the steps of the 5G offshore wind farm intelligent inspection and control method described above.

[0147] The technical solution of the present invention has been described in conjunction with the specific experimental procedures shown in the accompanying drawings. However, the scope of protection of the present invention is not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions resulting from such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A method for intelligent inspection and control of 5G offshore wind farms, characterized in that, Includes the following steps: S1. Collect real-time fault prediction data of wind farms, including wind turbine status parameters and environmental parameters, through multi-source heterogeneous sensors deployed at offshore wind farms. S2. Based on the 5G communication network, the fault prediction data is transmitted to the edge computing node in real time, and the edge computing node performs preprocessing operations on the fault prediction data. S3. Upload the preprocessed real-time prediction data to the cloud analysis platform, and use a multimodal long short-term memory network fault prediction model with an attention mechanism to extract key time series features and generate a fault risk membership profile for each wind turbine based on a fuzzy clustering algorithm. S4. Based on the risk membership degree output by the cloud model, construct a multi-objective optimization function and use a genetic algorithm to generate the globally optimal unmanned inspection path and scheduling strategy. S5. Based on historical operating conditions and fault feedback data during the actual operation of the wind turbine, the fault prediction model is updated offline on a regular basis.

2. The method according to claim 1, characterized in that: In step S1, the fan status parameters are obtained through vibration sensors, temperature sensors, blade speed sensors, and sound acquisition devices.

3. The method according to claim 1, characterized in that: In step S1, environmental parameters are acquired using an anemometer, temperature and humidity sensors, and aerosol lidar.

4. The method according to claim 1, characterized in that: Step S2 specifically includes the following steps: S21. Perform data cleaning operations on the operating status parameters and environmental information of each wind turbine, including outlier detection, missing value completion and format standardization. S22. The original sequence is reconstructed by using a sliding time window mechanism, dividing it into fixed-length time segments for use as input for time series modeling and prediction in the cloud.

5. The method according to claim 1, characterized in that: Step S3 specifically includes the following steps: S31. Input Construction: After the sliding window reconstruction of the wind turbine status and environmental data is completed at the edge computing node, the obtained input samples are organized into two synchronously aligned time series matrices: Wind turbine state parameter matrix, Where T represents the length of the time window, These include: fan vibration frequency, gearbox temperature, fan blade speed, and sound decibels; Environmental parameter matrix Where T represents the length of the time window, This includes: wind speed, air temperature, air humidity, and sea salt particle concentration; S32 and LSTM timing modeling were performed, using two LSTM networks respectively. , After processing, we get: Hidden sequence of state parameters: , Hidden sequence of environmental parameters: ; S33. Attention score calculation: For each time step of the hidden state in the hidden sequence of each of the two channels... Calculate attention score : In the formula, , , These are trainable parameter vectors and weight matrices; S34. Attention weight normalization and context aggregation: The score vector is normalized to obtain the weight coefficients. : This is used to perform weighted aggregation of the sequences and calculate the channel context feature vector: The state parameter channel context vectors are obtained respectively. With environmental parameter channel context vector ; S35. Feature fusion and risk prediction output: The two attention context vectors are concatenated to form a feature fusion vector. The input is fed into the fully connected layer to generate a fault probability prediction value. And normalized using the Sigmoid activation function; S36. Construction of fuzzy risk profiles, using the fuzzy C-means clustering algorithm (FCM) to assess failure probabilities. And process the feature fusion vector: Set cluster center set These correspond to the normal state, the potential hidden danger state, and the serious fault state, respectively. Calculate the membership degree of each wind turbine node i to risk category j. This forms the risk membership matrix U, which serves as the input for downstream regulatory weights. In the formula, Let be the feature vector of wind turbine i, and m be the fuzzy weighted index.

6. The method according to claim 5, characterized in that: Step S32 is as follows: S321. Set the hyperparameter set: In the formula, These represent the number of LSTM cells in the wind turbine state network and the offshore environment network, respectively. For learning rate, The L2 regularization coefficient is... Batch size; S322. Define the binary cross-entropy loss function on the validation set as the optimization objective; S323. Establish hyperparameters using a Gaussian process surrogate model. To determine the optimal hyperparameter combination based on the relationship between the validation loss and the expected improvement (EI) or probabilistic improvement (PI) strategies, the following approach can be used: 。 7. The method according to claim 1, characterized in that: Step S4 specifically includes the following steps: S41. Problem coding: Constructing chromosomes for the genetic algorithm using integer coding, chromosome sequence. The order in which the unmanned inspection equipment visits the fans is represented. Number the wind turbine; S42. Define a fitness function F with the goal of maximizing risk capture benefits and minimizing inspection costs: In the formula, These are the weighting coefficients. Let the membership degree of fan k belong to the severe fault class be denoted as . The time decay factor, The time to access wind turbine k, The Euclidean distance between the two wind turbines; S43. Genetic evolution: Perform selection, crossover, and mutation operations on the initial population. Selection operation: Use the roulette wheel algorithm to select individuals with high fitness; Crossover operation: The sequential crossover method OX is used to generate the child path while preserving the risk priority sequence characteristics of the parent generation; Mutation operation: Randomly swap the positions of two genes in a chromosome sequence to escape local optima; S44. Strategy Output: When the number of iterations reaches a preset value, output the chromosome sequence with the highest fitness as the final unmanned inspection path.

8. A 5G intelligent inspection and control system for offshore wind farms, characterized in that, Includes the following modules: The sensing and acquisition module is used to collect wind power equipment operating status parameters and environmental information in real time through multi-source heterogeneous sensors deployed on the offshore wind turbine and its operating environment. The network transmission module is used to transmit the sensed data to the edge computing node at high speed and stably through the 5G communication module, and supports edge access gateway, clock synchronization and link QoS guarantee functions; The edge preprocessing module is used to clean, remove outliers, fill in missing data, convert formats, and reconstruct sliding time windows for received data at edge computing nodes, and then upload the structured time series data to the cloud platform. The cloud-based analytics module is used to perform multimodal LSTM modeling and attention mechanism fusion, and includes fuzzy clustering units to generate membership matrices of wind turbines for different risk levels. The scheduling and control module integrates a genetic algorithm optimization engine, which is used to calculate the globally optimal inspection path and issue control commands based on the risk membership matrix and the geographical distribution of the wind field. The application business module is used to display a visual map of wind farm fault risks, configure remote operation and maintenance tasks, and provide operation and maintenance support functions such as remote feedback monitoring and historical record management.

9. An electronic device, comprising a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus, characterized in that: The memory is used to store computer programs; A processor, wherein the processor is configured to execute a computer program stored in a memory to implement the steps of the 5G offshore wind farm intelligent inspection and control method as described in any one of claims 1-7.

10. A non-volatile storage medium, characterized in that: The non-volatile storage medium stores an executable program, which, when executed by a processor, implements the steps of the 5G offshore wind farm intelligent inspection and control method as described in any one of claims 1-7.

Images