An offshore wind turbine operation data analysis and prediction system

Abstract

The application relates to the technical field of wind turbine data analysis, and discloses a system for analyzing and predicting offshore wind turbine operation data. The system comprises a marine environment data integration module, which collects data to generate a multi-source spatiotemporal synchronous data set; a multi-modal feature fusion module, which extracts cross-modal correlation features to generate a high-dimensional fusion feature tensor; a dynamic fault prediction module, which constructs a bidirectional gated recurrent network model to predict the degradation probability and residual life of key components of equipment; and a self-adaptive optimization control module, which constructs a multi-objective dynamic programming model to optimize the operation strategy of the wind turbine. In addition, the system is also provided with a feedback correction module to correct the parameters of the prediction model, and a virtual sensor module based on a physical information neural network to monitor the tower stress and diagnose sensor faults. The system realizes comprehensive monitoring, accurate fault prediction and optimized control of offshore wind turbines, effectively improves the operation efficiency, reliability and safety of the wind turbines, and reduces the operation and maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of wind turbine data analysis technology, specifically to a system for analyzing and predicting offshore wind turbine operation data. Background Technology

[0002] With the continuous growth of global demand for clean energy, offshore wind power, with its abundant resources and stable wind speeds, has become an important development direction in the renewable energy field. Offshore wind turbines operate in complex marine environments and face numerous challenges, placing extremely high demands on their operational stability, reliability, and safety, and prompting continuous technological innovation.

[0003] The marine environment is complex and changeable, with harsh conditions such as strong winds, giant waves, and salt spray posing significant challenges to the operation of offshore wind turbines. Strong winds can cause excessive loads on turbine blades, leading to fatigue damage or even breakage; the impact of giant waves can affect the stability of the turbine tower, increasing the risk of structural damage; and salt spray is corrosive, accelerating the aging of equipment components and seriously threatening the safe operation of wind turbines.

[0004] Current wind turbine operation data monitoring has many shortcomings. Existing monitoring systems collect a limited number of data types, making it difficult to comprehensively reflect the operating status of wind turbines and marine environmental information. Furthermore, the accuracy and synchronization of the data are poor; data collected by different sensors exhibit temporal and spatial discrepancies, failing to provide reliable data for subsequent analysis. For example, the sampling frequencies of some wind speed sensors and vibration sensors are inconsistent, leading to inaccurate data matching and affecting the assessment of equipment operating status.

[0005] Insufficient fault prediction capability is one of the key problems currently facing offshore wind turbines. Traditional fault prediction methods are mostly based on single data points or simple models, failing to fully explore the potential correlations between multi-source data. In the complex marine environment, equipment failures are often the result of multiple factors working together, making it difficult for single data points to accurately predict the timing and extent of failures. This makes it difficult for maintenance personnel to take timely and effective maintenance measures, leading to prolonged equipment downtime and increased maintenance costs.

[0006] The control strategies for wind turbines need optimization. Existing control strategies are mostly based on fixed rules or experience, failing to fully consider the dynamic changes in the marine environment. Under different wind speeds and wave conditions, the power output, yaw angle, and pitch rate of the wind turbine cannot be optimally adjusted, resulting in low power generation efficiency and increased mechanical fatigue wear on the equipment. For example, when wind speed changes abruptly, the wind turbine cannot adjust the blade pitch angle in time, causing a reduction in wind energy utilization efficiency and subjecting the equipment to unnecessary shocks.

[0007] Furthermore, the sensor fault detection and handling mechanisms for offshore wind turbines are inadequate. Sensors are prone to failure in the harsh marine environment, and sensor failure can lead to data loss or errors, affecting the normal operation of the entire monitoring and control system. Existing fault detection methods are slow to respond and cannot detect and handle sensor faults in a timely manner, thus reducing the reliability of the system. Summary of the Invention

[0008] The purpose of this invention is to provide a system for analyzing and predicting the operation data of offshore wind turbines to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a system for analyzing and predicting the operation data of offshore wind turbines, the system comprising:

[0010] The marine environmental data integration module is used to collect operating data of wind turbines and marine environmental parameters to generate multi-source spatiotemporal synchronous datasets.

[0011] The multimodal feature fusion module is used to map the multi-source spatiotemporal synchronous dataset to a graph neural network architecture, and extract cross-modal correlation features of equipment status, wave load and salt spray concentration through node embedding and edge weight dynamic allocation algorithms to generate a high-dimensional fusion feature tensor.

[0012] The dynamic fault prediction module is used to construct a bidirectional gated recurrent network model based on a spatiotemporal attention mechanism, perform time-series dependency modeling on the high-dimensional fusion feature tensor, and output the degradation probability and remaining lifetime prediction values ​​of key components of the device.

[0013] The adaptive optimization control module is used to construct a multi-objective dynamic programming model based on the predicted values, and to optimize the wind turbine power output, yaw angle and pitch rate using a non-dominated sorting genetic algorithm to generate an anti-disturbance operation strategy.

[0014] Preferably, the multi-source spatiotemporal synchronization dataset is subjected to denoising and alignment processing based on wavelet transform and adaptive filtering algorithms, specifically including:

[0015] Vibration signals, wind speed and sea temperature sequences are extracted from multi-source spatiotemporal synchronous data, and high-frequency noise components and low-frequency trend components are generated by multi-resolution wavelet decomposition.

[0016] An adaptive filter is designed using the minimum mean square error criterion to dynamically suppress high-frequency components.

[0017] A timestamp alignment algorithm is used to interpolate and synchronize the denoised low-frequency components with the device status data.

[0018] Preferably, the generation of the high-dimensional fusion feature tensor includes:

[0019] Define device nodes, environment nodes, and load nodes as vertices of the graph structure, and construct the device-environment-load association matrix;

[0020] A graph attention mechanism is used to aggregate features of adjacent nodes to generate node-level spatiotemporal embedding vectors.

[0021] The strength of the association between nodes is optimized by using a dynamic edge weight update strategy, and the output is a feature tensor containing nonlinear coupling relationships.

[0022] Preferably, the construction of the bidirectional gated recurrent network model based on the spatiotemporal attention mechanism includes:

[0023] A bidirectional GRU network is used in the time dimension to capture the forward and backward dependencies of device states;

[0024] A spatial attention layer is introduced to calculate the weight distribution of data from each sensor, focusing on abnormal signals of key components;

[0025] By using a time attention layer to filter feature segments in historical sequences that are strongly correlated with the current fault, a time series curve of degradation probability is generated.

[0026] Preferably, the construction of the multi-objective dynamic programming model includes:

[0027] The objective function is defined as a weighted combination of maximizing power generation efficiency, minimizing mechanical fatigue loss, and minimizing the sum of squared yaw errors;

[0028] Set the constraints as the maximum allowable pitch angle, rated power threshold, and safe threshold range of wave impact force;

[0029] An elite retention strategy and a crowding comparison operator are used to optimize the Pareto solution set and screen for perturbation-resistant operating strategies.

[0030] Preferably, the system further includes:

[0031] The feedback correction module, used in the sliding time window-based incremental learning framework, combines Kalman filtering and Bayesian inference algorithms to perform online correction of the prediction model parameters, generating an adaptively updated fault prediction model; specifically, it includes:

[0032] Real-time data streams are divided based on sliding windows, and the prediction error sequence within the window is extracted as a feedback signal.

[0033] The Kalman gain matrix is ​​designed to recursively update the network weights, reducing reliance on historical data.

[0034] The model hyperparameters are adjusted using the Bayesian posterior probability distribution to achieve adaptive calibration of the prediction model.

[0035] Preferably, the multi-source spatiotemporal synchronization dataset includes:

[0036] Integrate unit electrical parameters collected by the SCADA system, vibration data from fiber optic sensors, and wave spectrum data provided by meteorological buoys;

[0037] Tensor completion algorithm is used to jointly impute missing segments of heterogeneous data to construct a complete spatiotemporal data cube;

[0038] The dimensional differences are eliminated by combining Z-score standardization and range normalization strategies.

[0039] Preferably, the system further includes:

[0040] A virtual sensor module based on physical information neural network is constructed to estimate the tower stress distribution in real time, which cannot be directly measured.

[0041] The estimated results are compared with the measured data to generate a residual signal, which triggers sensor fault diagnosis and data source switching commands.

[0042] Preferably, the present invention further includes an electronic device, the device comprising:

[0043] At least one processor; and a memory communicatively connected to said at least one processor;

[0044] The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform operations such as those described above based on the offshore wind turbine operation data analysis and prediction system.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] In terms of data processing, the marine environmental data integration module integrates wind turbine operation data with marine environmental parameters to generate a multi-source spatiotemporal synchronous dataset. A tensor completion algorithm is used to jointly impute missing segments of heterogeneous data, constructing a complete spatiotemporal data cube to ensure data integrity. A combination of Z-score standardization and range normalization is employed to eliminate dimensional differences, improving data quality and laying a solid foundation for subsequent accurate analysis. The fusion of multi-source data makes the description of wind turbine operation status more comprehensive and accurate, avoiding analytical biases caused by missing or inconsistent data.

[0047] The multimodal feature fusion module employs a graph neural network architecture to extract cross-modal correlation features of equipment status, wave load, and salt spray concentration, generating a high-dimensional fused feature tensor. This process constructs a correlation matrix by defining different types of nodes, aggregates features using a graph attention mechanism, and optimizes the correlation strength through a dynamic edge weight update strategy, fully exploring the complex nonlinear coupling relationships between multi-source data. Compared to traditional single-feature analysis methods, it can capture more key information hidden in the data, providing richer and more valuable data support for fault prediction and operational optimization.

[0048] The dynamic fault prediction module is based on a bidirectional gated recurrent network model with a spatiotemporal attention mechanism, which performs temporal dependency modeling on high-dimensional fused feature tensors. In the time dimension, the bidirectional GRU network captures the forward and backward dependencies of equipment states, enabling a more comprehensive understanding of the changing trends of equipment operating states over time. The spatial attention layer focuses on abnormal signals of key components, while the temporal attention layer filters feature fragments strongly correlated with the current fault, thereby accurately outputting the degradation probability and remaining life prediction values ​​of key equipment components. This prediction method significantly improves the accuracy and timeliness of fault prediction. Maintenance personnel can formulate maintenance plans in advance based on the prediction results, rationally allocate maintenance resources, effectively reduce downtime caused by sudden equipment failures, lower maintenance costs, and improve the reliability and availability of wind turbine units.

[0049] The adaptive optimization control module constructs a multi-objective dynamic programming model based on predicted values ​​and employs a non-dominated sorting genetic algorithm to optimize wind turbine power output, yaw angle, and pitch rate, generating an anti-disturbance operation strategy. This strategy aims to maximize power generation efficiency, minimize mechanical fatigue loss, and minimize the sum of squared yaw errors, while considering constraints such as the maximum allowable pitch angle, rated power threshold, and wave impact safety threshold. Through optimized control, the wind turbine can better adapt to changes in the marine environment, improving power generation efficiency, reducing mechanical fatigue loss, and extending equipment lifespan while ensuring safe and stable operation. For example, under different wind speeds and wave conditions, the turbine can automatically adjust its pitch angle and yaw angle to achieve efficient wind energy capture and reduce unnecessary equipment wear.

[0050] The feedback correction module utilizes an incremental learning framework based on a sliding time window, combining Kalman filtering and Bayesian inference algorithms to perform online correction of the prediction model parameters. It extracts the prediction error sequence using a sliding window as a feedback signal, recursively updates the network weights using the Kalman gain matrix to reduce reliance on historical data, and adjusts the model hyperparameters using the Bayesian posterior probability distribution, achieving adaptive calibration of the prediction model. This allows the model to adjust in real time according to changes in wind turbine operating conditions and the environment, maintaining high prediction accuracy and further enhancing the system's reliability and adaptability.

[0051] A virtual sensor module based on a physical information neural network was constructed to estimate the tower stress distribution in real time, which cannot be directly measured. By comparing this estimate with measured data, a residual signal is generated, triggering sensor fault diagnosis and data source switching commands. This module not only achieves effective monitoring of tower stress but also promptly detects sensor faults and switches data sources, ensuring the accuracy and continuity of system data. This improves the reliability and stability of the entire monitoring system, guaranteeing the safe and stable operation of the wind turbine. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of the working principle of the offshore wind turbine operation data analysis and prediction system described in this invention.

[0053] Figure 2 Flowchart for generating high-dimensional fusion feature tensors;

[0054] Figure 3 A flowchart illustrating the construction of a bidirectional gated recurrent network model based on a spatiotemporal attention mechanism;

[0055] Figure 4 This is a flowchart of the workflow for a virtual sensor module based on a physical information neural network. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see Figures 1-4 This invention provides a system for analyzing and predicting the operating data of offshore wind turbines. It aims to achieve monitoring of the operating status, fault prediction, and optimized control of offshore wind turbines through the collaborative work of multiple modules. The overall implementation scheme is as follows:

[0058] The system first collects operational data from the wind turbines and marine environmental parameters through a marine environmental data integration module. The operational data includes turbine electrical parameters and vibration data, while the marine environmental parameters include wave spectrum data and meteorological data. This data is integrated to generate a multi-source spatiotemporal synchronous dataset. This dataset integrates data from different sources and across different time and spatial dimensions, providing a comprehensive foundation for subsequent analysis.

[0059] The multimodal feature fusion module receives multi-source spatiotemporal synchronized datasets and maps them into a graph neural network architecture. Within this architecture, cross-modal correlation features between different modal data, such as equipment status, wave load, and salt spray concentration, are extracted through node embedding and dynamic edge weight allocation algorithms. These features are then processed to generate a high-dimensional fusion feature tensor, effectively fusing feature information from multiple sources and providing richer data support for subsequent predictions.

[0060] The dynamic fault prediction module constructs a bidirectional gated recurrent network model based on a spatiotemporal attention mechanism. This model performs temporal dependency modeling on the high-dimensional fused feature tensor, fully considering the correlation of data in time and space dimensions, and outputs the degradation probability and remaining life prediction values ​​of key components of the equipment. In this way, the failure conditions of key components of wind turbines can be predicted in advance, providing an important basis for maintenance decisions.

[0061] The adaptive optimization control module constructs a multi-objective dynamic programming model based on the predicted values ​​output by the dynamic fault prediction module. A non-dominated sorting genetic algorithm is used to optimize the wind turbine's power output, yaw angle, and pitch rate, generating an anti-disturbance operation strategy. This strategy enables the wind turbine to maintain stable operation in complex marine environments, improving power generation efficiency and reducing equipment losses.

[0062] The implementation of the present invention will be further described below with reference to Examples 1 to 6.

[0063] Example 1:

[0064] During system operation, data quality is crucial to the accuracy of subsequent analysis and prediction. Therefore, for multi-source spatiotemporal synchronized datasets, denoising and alignment processing are necessary, while ensuring the integrity and standardization of the dataset.

[0065] Multi-source spatiotemporal synchronous data includes vibration signals, wind speed, and seawater temperature sequences. Utilizing the multi-resolution characteristics of wavelet transform, these sequences are decomposed using multi-resolution wavelet decomposition. During wavelet decomposition, the data is divided into high-frequency noise components and low-frequency trend components. The high-frequency noise components primarily contain noise information from the data, while the low-frequency trend components retain the main trend characteristics of the data. For example, after wavelet decomposition, the high-frequency components of a vibration signal may contain minute interference noise generated during equipment vibration, while the low-frequency components reflect the overall trend of equipment vibration.

[0066] To dynamically suppress high-frequency noise, an adaptive filter is designed using the minimum mean square error (MSE) criterion. The principle of the MSE criterion is to continuously adjust the filter parameters to minimize the mean square error between the filter output and the desired output. In practical applications, the desired output can be determined by analyzing the original signal or by referencing other relevant data. The adaptive filter dynamically adjusts its parameters according to changes in the input signal, effectively suppressing noise in the high-frequency components and thus improving data quality.

[0067] After denoising, the low-frequency components and equipment status data need to be interpolated and synchronized. A timestamp alignment algorithm is used to process the denoised low-frequency components and equipment status data based on the timestamp information. For missing or out-of-synchronization data points in the time series, interpolation methods are used to supplement and synchronize them. For example, if wind speed data is missing at a certain moment, the wind speed value at that moment can be estimated using the wind speed data from before and after that moment through an interpolation algorithm, ensuring that all data are synchronized in time and providing an accurate data foundation for subsequent analysis and processing.

[0068] The multi-source spatiotemporal synchronous dataset integrates various heterogeneous data, including unit electrical parameters collected by the SCADA system, vibration data from fiber optic sensors, and wave spectrum data provided by meteorological buoys. Since various factors may lead to data loss during data acquisition, a tensor completion algorithm is employed to jointly impute missing segments of the heterogeneous data. Based on the low-rank property of tensors, the tensor completion algorithm estimates the value of missing data through analysis and learning of known data. For example, in a three-dimensional spatiotemporal data cube, if a certain type of data is missing at a certain time and location, the tensor completion algorithm can utilize relevant data from other times and locations to calculate the estimated value of the missing data, thereby constructing a complete spatiotemporal data cube.

[0069] To eliminate the dimensional differences between different data points, a combination of Z-score standardization and range normalization is employed. The formula for Z-score standardization is: Where x represents the original data, μ is the mean of the data, and σ is the standard deviation of the data. Z-score standardization transforms the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The formula for range normalization is: Where x represents the original data, x min x is the minimum value in the data. max The maximum value in the data is represented by the range normalization function, which maps the data to the interval [0,1]. This combination of methods effectively eliminates dimensional differences between different data sets, making different types of data comparable and facilitating subsequent analysis and model training.

[0070] Example 2:

[0071] In the multimodal feature fusion module, to construct a graph structure reflecting the relationships between data, device nodes, environment nodes, and load nodes are defined as vertices of the graph structure. For example, device nodes can represent various components of a wind turbine, such as blades, gearboxes, and generators; environment nodes can represent various factors in the marine environment, such as wind speed, wave height, and salt spray concentration; and load nodes are used to describe various loads acting on the wind turbine, such as wave loads and wind loads. Based on the actual relationships between these nodes, a device-environment-load correlation matrix is ​​constructed. The elements in the correlation matrix represent the connection relationship and correlation strength between nodes. If there is a direct correlation between two nodes, the value of the corresponding element is not 0, and the magnitude of the value reflects the strength of the correlation.

[0072] A graph attention mechanism is employed to aggregate features from neighboring nodes. The core idea of ​​this mechanism is to determine the importance of each node in the feature aggregation process by calculating attention weights between nodes. For each node, a corresponding attention weight is calculated based on the feature information of its neighboring nodes. For example, for a device node, it focuses on the features of its neighboring environment and payload nodes, and by calculating attention weights, it more effectively aggregates the features of nodes that have a greater impact on it. In the specific calculation process, some learnable parameters are used to adjust the calculation method of the attention weights, thereby generating more representative node-level spatiotemporal embedding vectors. These embedding vectors integrate the feature information of the node itself and its neighboring nodes, better reflecting the node's characteristics within the entire graph structure.

[0073] To further optimize the correlation strength between nodes, a dynamic edge weight update strategy is used to adjust the edge weights in the correlation matrix. During wind turbine operation, the correlation between equipment status, environmental factors, and loads is not fixed but dynamically adjusts with changes in time and operating conditions. The dynamic edge weight update strategy continuously adjusts the edge weights in the correlation matrix based on real-time data and model feedback. For example, when wind speed changes significantly, the blade load and equipment status of the wind turbine may be significantly affected. In this case, it is necessary to increase the edge weights between the wind speed node and related equipment and load nodes to highlight this change in correlation. After dynamic edge weight updates, the output feature tensor contains more accurate nonlinear coupling relationships, providing more valuable information for subsequent fault prediction and analysis.

[0074] Example 3:

[0075] In the dynamic fault prediction module, constructing a bidirectional gated recurrent network model based on a spatiotemporal attention mechanism is the core of achieving accurate fault prediction.

[0076] In the time dimension, a bidirectional GRU (Gated Cyclic Unit) network is employed to capture the forward and backward dependencies of equipment states. The bidirectional GRU network consists of a forward GRU and a backward GRU. The forward GRU starts from the beginning of the data sequence and processes the data sequentially at each time point, learning the forward time dependencies of the data; the backward GRU starts from the end of the data sequence and processes the data in reverse, learning the backward time dependencies of the data. For example, when monitoring the vibration data of wind turbine blades, the forward GRU can learn the trend of blade vibration changes over time, while the backward GRU can capture the impact of possible abnormal changes in the vibration data in later stages. By concatenating the outputs of the forward and backward GRUs, the forward and backward dependencies of equipment states in the time dimension can be comprehensively captured, providing richer time-series information for subsequent fault analysis.

[0077] A spatial attention layer is introduced to calculate the weight distribution of data from various sensors. In wind turbines, multiple sensors are installed to monitor information from different locations and parameters. However, the importance of different sensor data for fault prediction is not the same. The spatial attention layer calculates the weight of each sensor data point in fault prediction based on its characteristics. For example, when judging a gearbox fault, the gearbox temperature sensor data may have a relatively higher weight than the tower vibration sensor data. In this way, the spatial attention layer can focus on abnormal signals from critical components, highlighting sensor data that has a significant impact on fault prediction and improving the accuracy of fault prediction.

[0078] The temporal attention layer is used to filter feature segments in the historical data sequence that are strongly correlated with the current fault. During equipment operation, not all segments in the historical data have an equal degree of correlation with the current fault. The temporal attention layer analyzes the historical data sequence based on the current data and the model's learning results, calculating the correlation between each historical segment and the current fault—the attention weight. Then, based on these weights, it filters out feature segments that are strongly correlated with the current fault. For example, when predicting generator faults, the temporal attention layer might find a strong correlation between vibration data following a sudden change in generator load over a past period and the current generator fault, thus filtering out this data segment. In this way, the generated degradation probability time-series curve can more accurately reflect the actual degradation of key equipment components, providing a reliable basis for predicting the remaining life of the equipment.

[0079] Example 4:

[0080] In the adaptive optimization control module, the objective function is defined as a weighted combination of maximizing power generation efficiency, minimizing mechanical fatigue loss, and minimizing the sum of squared yaw errors. Maximizing power generation efficiency aims to increase the power generation capacity of the wind turbine and obtain more electricity revenue. Minimizing mechanical fatigue loss helps extend the service life of the equipment and reduce maintenance costs. Minimizing the sum of squared yaw errors enables the wind turbine to better track the wind direction, improving wind energy capture efficiency. The objective function can be expressed as: Where J is the objective function value, w1, w2, and w3 are the weighting coefficients corresponding to power generation efficiency, mechanical fatigue loss, and the sum of squared yaw error, respectively, and P is the objective function value. actual P represents the actual power generation. max For maximum power generation, F fatigue F represents the mechanical fatigue loss value. threshold e is the fatigue loss threshold. yaw Let yaw error be denoted by , and n be the number of sampling points. Weighting coefficients w1, w2, and w3 are set according to actual needs and importance to balance the relationships between different targets.

[0081] The constraints are set as the maximum permissible pitch angle, the rated power threshold, and the safe threshold range for wave impact. The maximum permissible pitch angle limits the blade adjustment range to prevent excessive blade adjustment that could damage the equipment. The rated power threshold ensures that the wind turbine operates within a safe and stable power range, avoiding overload. The safe threshold range for wave impact takes into account the impact of waves on the wind turbine in the marine environment, ensuring that the wind turbine does not exceed safe limits when subjected to wave impact. For example, the maximum permissible pitch angle might be set to [0, 90°], and the rated power threshold to [0, P]. rated The safe threshold range for wave impact force is determined based on the design of the wind turbine and the actual marine environment. In actual operation, the operating parameters of the wind turbine must meet these constraints; otherwise, corresponding protection mechanisms or adjustment strategies will be triggered.

[0082] An elite retention strategy and a crowding comparison operator are employed to optimize the Pareto solution set and screen for disturbance-resistant operating strategies. The Pareto solution set contains multiple non-dominated solutions that satisfy different objective trade-offs. The elite retention strategy refers to directly retaining the best individuals from the current generation to the next generation during the iteration process of the genetic algorithm, preventing the loss of excellent individuals during evolution. The crowding comparison operator is used to measure the degree of crowding among individuals in the Pareto solution set, selecting those solutions that are more evenly distributed and far away from other individuals. For example, when calculating the crowding degree of an individual, it can be measured by calculating the distance between the individual and its neighboring individuals in the target space. By combining these two methods, a better disturbance-resistant operating strategy can be screened from the Pareto solution set, enabling wind turbine units to improve power generation efficiency and ensure safe and stable operation of the equipment while reducing various losses in complex marine environments.

[0083] Example 5:

[0084] To improve the accuracy and adaptability of the prediction model, the system incorporates a feedback correction module. This module uses an incremental learning framework based on a sliding time window, combined with Kalman filtering and Bayesian inference algorithms, to perform online correction of the prediction model parameters.

[0085] The real-time data stream is divided using a sliding window. Real-time data refers to various monitoring data continuously generated during wind turbine operation, such as equipment status data and environmental data. The size of the sliding window is set according to actual conditions; for example, it can be set to include data from the past 10 minutes. Within each time window, a prediction error sequence is extracted as a feedback signal. The prediction error sequence refers to the sequence of differences between the model's predicted values ​​and the actual measured values. By analyzing these errors, the prediction deviation of the model can be understood, providing a basis for model parameter correction. For example, if there is a significant deviation between the model's predicted generator temperature and the actual measured temperature within a certain window, this deviation value will be included in the prediction error sequence.

[0086] The Kalman gain matrix is ​​designed to recursively update the network weights. Kalman filtering is a commonly used state estimation method that optimizes the estimation of the system state by continuously fusing new measurement data with previous estimation results. In this system, the network weights are considered as the system state, and the Kalman gain matrix is ​​used to adjust the weight updates. The calculation of the Kalman gain matrix is ​​based on factors such as the current measurement noise and the system noise covariance. For example, when the measurement data has high noise, the Kalman gain matrix will be adjusted accordingly, making the weight updates more conservative and reducing the impact of noise on the weight updates. This recursive update method reduces the dependence on historical data, allowing the model to adapt to new data changes more quickly and improving the model's real-time performance and accuracy.

[0087] The model hyperparameters are adjusted using Bayesian posterior probability distribution. Bayesian inference is a probability-based reasoning method that uses prior knowledge and new observational data to update the probability estimates of unknown parameters. In this system, model hyperparameters are treated as unknown parameters, and these hyperparameters are adjusted using Bayesian posterior probability distribution. For example, for hyperparameters such as the learning rate and the number of hidden layer nodes in a neural network, a prior probability distribution is first set based on experience or previous experiments. Then, combined with new observational data (such as prediction error sequences), the posterior probability distribution is calculated using Bayes' theorem. Based on the posterior probability distribution, the optimal hyperparameter values ​​are selected to achieve adaptive calibration of the prediction model. In this way, the model can automatically adjust hyperparameters under different operating conditions, improving the model's predictive performance and adaptability.

[0088] Example 6:

[0089] In the system, a virtual sensor module based on physical information neural network is constructed to estimate the tower stress distribution in real time, which cannot be directly measured, and to realize sensor fault diagnosis and data source switching by comparing with measured data.

[0090] Virtual sensors are constructed based on physical information neural networks (PINs). A PSN is a model that combines physical principles with neural networks. When estimating the stress distribution of a tower, a corresponding physical model is first established based on the tower's physical structure and mechanical principles. For example, based on the tower's material properties, geometry, and load conditions, theoretical stress values ​​for the tower at different locations and under different operating conditions are calculated using mechanical formulas. Then, this information from the physical model is integrated into the neural network. The neural network learns from a large amount of actual operational data, including environmental data related to tower stress (such as wind speed and direction) and equipment operating data (such as blade speed and generator power), to adjust its parameters and more accurately estimate the tower stress distribution. Compared to traditional neural networks, PSNs utilize the prior knowledge of physical models, enabling them to improve the accuracy and generalization ability of models even with limited data.

[0091] The tower stress distribution estimated by a virtual sensor is compared with measured data to generate a residual signal. The measured data is acquired through sensors installed on the tower, such as strain gauge sensors. The residual signal is the difference between the estimated and measured values. If the residual signal fluctuates within a certain range, it indicates that the virtual sensor's estimation results are in good agreement with the actual situation, and the sensor is functioning normally. If the residual signal exceeds a set threshold range, it indicates a possible sensor malfunction or other abnormality. For example, if the absolute value of the residual signal exceeds a certain set threshold for multiple consecutive moments, it can be determined that the sensor may be malfunctioning.

[0092] Once an anomaly in the residual signal is detected, the system triggers sensor fault diagnosis and data source switching commands. During fault diagnosis, the characteristics, trends, and other related data of the residual signal are further analyzed to determine the type and possible cause of the fault. For example, by analyzing the rate of change and periodicity of the residual signal, it can be determined whether the anomaly is due to a hardware fault in the sensor itself or external interference. If a sensor fault is determined, the system automatically switches the data source to obtain data from a backup sensor or other relevant data sources to ensure the normal operation of the system and continuous monitoring of tower stress distribution. Simultaneously, the system records fault information for subsequent maintenance and analysis. This approach improves the reliability and stability of the system, ensuring that sensor faults can be detected and addressed promptly during wind turbine operation.

[0093] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0094] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for predicting based on operational data analysis of offshore wind turbines, characterized in that, The system comprises: An ocean environment data integration module for collecting operation data of a wind turbine and ocean environment parameters, and generating a multi-source spatio-temporal synchronous data set; A multi-modal feature fusion module for mapping the multi-source spatio-temporal synchronous data set into a graph neural network architecture, extracting cross-modal associated features of equipment state, wave load and salt mist concentration through a node embedding and edge weight dynamic allocation algorithm, and generating a high-dimensional fusion feature tensor; A dynamic fault prediction module for constructing a bidirectional gated recurrent network model based on a spatio-temporal attention mechanism, modeling the high-dimensional fusion feature tensor in a time sequence dependence manner, and outputting a degradation probability and a remaining life prediction value of a key component of the equipment; An adaptive optimization control module for constructing a multi-objective dynamic programming model according to the prediction value, optimizing wind turbine power output, yaw angle and variable pitch rate by using a non-dominated sorting genetic algorithm, and generating an anti-disturbance operation strategy; A feedback correction module for generating an adaptively updated fault prediction model based on a sliding time window incremental learning framework, combining Kalman filtering and Bayesian inference algorithms to perform online correction on prediction model parameters; specifically comprising: Dividing real-time data streams based on a sliding window, and extracting a prediction error sequence in the window as a feedback signal; Designing a Kalman gain matrix to recursively update network weights, and reducing historical data dependence; Adjusting model hyperparameters using a Bayesian posterior probability distribution to achieve adaptive calibration of the prediction model; Constructing a virtual sensor module based on a physical information neural network to estimate tower stress distribution that cannot be directly measured in real time; Comparing the estimation results with measured data to generate a residual signal, and triggering sensor fault diagnosis and data source switching instructions.

2. The system for analyzing and predicting based on operational data of offshore wind turbines as claimed in claim 1 wherein, Based on wavelet transform and adaptive filtering algorithm, the multi-source spatio-temporal synchronous data set is denoised and aligned, specifically comprising: Extracting vibration signals, wind speed and seawater temperature sequences in the multi-source spatio-temporal synchronous data, and generating high-frequency noise components and low-frequency trend components through multi-resolution wavelet decomposition; Designing an adaptive filter based on the least mean square error criterion to dynamically suppress high-frequency components; Using a timestamp alignment algorithm to interpolate and synchronize the denoised low-frequency components and equipment state data.

3. The system for analyzing and predicting based on operational data of offshore wind turbines as claimed in claim 1 wherein, The high-dimensional fusion feature tensor is generated, comprising: Defining device nodes, environment nodes and load nodes as graph structure vertices to construct a device-environment-load correlation matrix; Using a graph attention mechanism to aggregate features of adjacent nodes to generate node-level spatio-temporal embedding vectors; Optimizing the correlation strength between nodes through a dynamic edge weight update strategy to output a feature tensor containing nonlinear coupling relationships.

4. The system for analyzing and predicting based on operational data of offshore wind turbines of claim 1, wherein, The bidirectional gated recurrent network model based on the spatio-temporal attention mechanism is constructed, comprising: Using a bidirectional GRU network to capture the forward and backward dependence of the equipment state in the time dimension; Introducing a spatial attention layer to calculate the weight distribution of each sensor data and focus on abnormal signals of key components; Through a time attention layer, filtering feature fragments in the historical sequence that are strongly related to the current fault to generate a degradation probability time sequence curve.

5. The system for analyzing and predicting based on operational data of offshore wind turbines as claimed in claim 1 wherein, The multi-objective dynamic programming model is constructed, comprising: The objective function is defined as a weighted combination of maximizing power generation efficiency, minimizing mechanical fatigue loss and minimizing yaw error square sum; The constraint conditions are set as the maximum allowable pitch angle, the rated power threshold and the safety threshold range of wave impact force; The elitist strategy and crowding distance comparison operator are used to optimize the Pareto solution set and screen the anti-disturbance operation strategy.

6. The system for analyzing and predicting based on operational data of offshore wind turbines as claimed in claim 1 wherein, The multi-source spatio-temporal synchronous data set comprises: The electrical parameters of the generating units collected by the SCADA system, the vibration data of the fiber-optic sensors and the wave spectrum data provided by the weather buoy are integrated; A tensor completion algorithm is used to jointly interpolate the missing sections of the heterogeneous data to construct a complete spatio-temporal data cube; The dimension difference is eliminated by a combination strategy of Z-score standardization and range normalization.

7. An electronic device, comprising: The system comprises: at least one processor; and a memory connected in communication with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the operations of the system according to any one of claims 1 to 6.

Images