Operation mode switching control method and device for offshore wind turbine generator

Abstract

The invention relates to the technical field of data processing, and discloses an operation mode switching control method and device for an offshore wind turbine generator. The method comprises the following steps: carrying out model classification on historical operation data of a unit to obtain operation mode classification data and a characteristic parameter library; performing reinforcement learning analysis on the current state evaluation data and the short-term prediction data of the unit to obtain a target operation mode; performing fuzzy control processing on the target operation mode to obtain a mode switching strategy and a parameter adjustment sequence; performing multi-objective optimization and cooperative control analysis on the mode switching strategy and the parameter adjustment sequence to obtain accurate control signals of each actuator in the target offshore wind turbine generator; and generating an operation mode switching control strategy according to the accurate control signal of each actuator, and performing mode switching control on the target offshore wind turbine generator through the mode switching control strategy. According to the invention, the efficiency and accuracy of the operation mode switching control of the offshore wind turbine generator are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to an operation mode switching control method and device for an offshore wind turbine. Background Art

[0002] Existing control methods for switching operating modes of offshore wind turbines mainly rely on preset fixed thresholds and simple logical judgments. These methods usually make mode switching decisions based on single parameters such as wind speed and power output, and use fixed control parameters to adjust actuators. Some advanced control systems have introduced intelligent algorithms such as fuzzy control and neural networks to improve the flexibility and adaptability of control.

[0003] However, these existing technologies often find it difficult to make timely and accurate mode switching decisions when faced with complex and changeable offshore environments. Fixed thresholds are difficult to adapt to the optimal operating state under different sea conditions, and simple logical judgments cannot fully consider the coupling relationship between multiple parameters. At the same time, the lack of in-depth mining and learning of historical data makes it difficult for control strategies to be continuously optimized and adapt to long-term changes such as equipment aging. In addition, when performing mode switching, the existing methods are not precise enough in the coordinated control of the actuators, which can easily cause unit vibration and load fluctuations, affecting equipment life and power generation efficiency. Summary of the invention

[0004] The present application provides an operation mode switching control method and device for an offshore wind turbine generator set, which are used to improve the efficiency and accuracy of the operation mode switching control of the offshore wind turbine generator set.

[0005] In a first aspect, the present application provides an operation mode switching control method for an offshore wind turbine group, the operation mode switching control method for an offshore wind turbine group comprising: real-time acquisition and preprocessing of the operation parameters of the target offshore wind turbine group to obtain the current state evaluation data and short-term prediction data of the unit; model classification of the pre-acquired historical operation data of the unit to obtain operation mode classification data and a feature parameter library; reinforcement learning analysis of the current state evaluation data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode; fuzzy control processing of the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence; multi-objective optimization and collaborative control analysis of the mode switching strategy and the parameter adjustment sequence to obtain precise control signals of each actuator in the target offshore wind turbine group; generating an operation mode switching control strategy according to the precise control signals of each actuator, and performing mode switching control on the target offshore wind turbine group through the mode switching control strategy.

[0006] In combination with the first aspect, in a first implementation method of the first aspect of the present application, the real-time collection and preprocessing of the operating parameters of the target offshore wind turbine unit to obtain the unit's current state evaluation data and short-term prediction data include: real-time monitoring of the wind speed, wind direction, turbulence intensity, wave height and tidal conditions of the target offshore wind turbine unit to obtain environmental parameter data; real-time collection of the power output, blade pitch angle, hub speed, main bearing temperature, gearbox vibration and generator current of the target offshore wind turbine unit to obtain unit operating parameter data; denoising the environmental parameter data and the unit operating parameter data to obtain preprocessed environmental parameter data and unit operating parameter data; time-scaling the preprocessed environmental parameter data Sequence analysis is performed to obtain short-term environmental prediction data; statistical analysis is performed on the pre-processed unit operating parameter data to obtain the unit current operating status data; regression analysis is performed on the short-term environmental prediction data through a support vector regression algorithm to obtain environmental parameter prediction values; the unit current operating status data is evaluated through a fuzzy inference algorithm to obtain the unit current status evaluation result; data fusion is performed on the environmental parameter prediction value and the unit current status evaluation result to obtain comprehensive status prediction data; the comprehensive status prediction data is filtered through a Kalman filter algorithm to obtain optimized status prediction data; threshold judgment is performed on the optimized status prediction data to obtain the unit current status evaluation data and the short-term prediction data.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present application, the model classification of the pre-acquired historical operation data of the unit to obtain operation mode classification data and a feature parameter library includes: data cleaning of the historical operation data of the unit to obtain a valid historical data set; time series segmentation of the valid historical data set to obtain multiple time window data; feature extraction of the multiple time window data by a principal component analysis algorithm to obtain reduced dimension feature data; cluster analysis of the reduced dimension feature data to obtain a preliminary operation mode category; boundary optimization of the preliminary operation mode category by a support vector machine algorithm to obtain an optimized operation mode category; labeling of the optimized operation mode category to obtain labeled operation mode data; rule extraction of the labeled operation mode data by a decision tree algorithm to obtain an operation mode classification rule; cross-validation of the operation mode classification rule to obtain a verified classification rule; fuzzy boundary processing of the verified classification rule by a fuzzy C-means algorithm to obtain a fuzzy classification rule; association analysis of the fuzzy classification rule and the reduced dimension feature data to obtain the operation mode classification data and the feature parameter library.

[0008] In combination with the first aspect, in a third implementation of the first aspect of the present application, the current state evaluation data of the unit and the short-term prediction data are subjected to reinforcement learning analysis through a preset deep neural network to obtain a target operating mode, including: data fusion of the current state evaluation data of the unit and the short-term prediction data to obtain comprehensive state data; feature extraction of the comprehensive state data through an autoencoder algorithm to obtain a compressed feature vector; normalization of the compressed feature vector to obtain standardized feature data; state space mapping of the standardized feature data through a deep Q network algorithm to obtain a state representation; discretization of the state representation in action space to obtain a candidate action set; exploration of the candidate action set through a Monte Carlo tree search algorithm to obtain an optimal action sequence; reward function evaluation of the optimal action sequence to obtain an action value estimate; gradient update of the action value estimate through a policy gradient algorithm to obtain optimized policy parameters; Bayesian inference of the optimized policy parameters to obtain an operating mode probability distribution; decision making on the operating mode probability distribution through the maximum a posteriori probability criterion to obtain the target operating mode.

[0009] In combination with the first aspect, in a fourth implementation method of the first aspect of the present application, the target operating mode is fuzzy controlled by a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence, including: extracting features from the target operating mode to obtain a mode feature vector; fuzzifying the mode feature vector to obtain a fuzzy input set; matching the fuzzy input set with a fuzzy inference engine to obtain an activation rule set; performing fuzzy inference operations on the activation rule set to obtain a fuzzy output set; defuzzifying the fuzzy output set by a centroid method to obtain a preliminary switching strategy; performing a safety assessment on the preliminary switching strategy to obtain a safe switching interval; optimizing the parameters of the safe switching interval by an adaptive fuzzy PID algorithm to obtain optimized switching parameters; arranging the optimized switching parameters in time series to obtain a parameter adjustment sequence; performing causal relationship analysis on the parameter adjustment sequence by a fuzzy cognitive map algorithm to obtain a switching influencing factor; and comprehensively processing the switching influencing factor and the parameter adjustment sequence to obtain the mode switching strategy and the parameter adjustment sequence.

[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, the mode switching strategy and the parameter adjustment sequence are subjected to multi-objective optimization and collaborative control analysis to obtain precise control signals of each actuator in the target offshore wind turbine, including: decoupling analysis of the mode switching strategy and the parameter adjustment sequence to obtain initial control targets of each actuator; setting constraints on the initial control targets of each actuator to obtain boundary conditions of the optimization problem; solving the optimization problem by a multi-objective genetic algorithm to obtain a Pareto optimal solution set; and performing hierarchical analysis on the Pareto optimal solution set. Obtain the weight coefficient of each target; evaluate the Pareto optimal solution set by a fuzzy comprehensive evaluation method to obtain the optimal solution; perform actuator allocation on the optimal solution to obtain the preliminary control signal of each actuator; perform feedforward optimization on the preliminary control signal of each actuator by a model predictive control algorithm to obtain an optimized control sequence; perform stability analysis on the optimized control sequence to obtain a stability evaluation result; perform robustness enhancement on the optimized control sequence by an H-infinity control algorithm to obtain a robust control signal; perform dynamic compensation on the robust control signal to obtain an accurate control signal of each actuator.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, the operation mode switching control strategy is generated according to the precise control signals of each actuator, and the target offshore wind turbine is subjected to mode switching control through the mode switching control strategy, including: arranging the precise control signals of each actuator in time sequence to obtain a control sequence; optimizing the control sequence through a dynamic programming algorithm to obtain an optimal control path; segmenting the optimal control path to obtain a stage control target; performing fuzzy reasoning on the stage control target through a fuzzy logic controller to obtain a fuzzy control rule; defuzzifying the fuzzy control rule to obtain a precise control instruction; enhancing the robustness of the precise control instruction through a sliding mode control algorithm to obtain an anti-disturbance control strategy; performing a safety assessment on the anti-disturbance control strategy to obtain a safety control boundary; online adjusting the control parameters within the safety control boundary through an adaptive control algorithm to obtain an adaptive control strategy; simulating and verifying the adaptive control strategy to obtain a control effect evaluation result; and fine-tuning the adaptive control strategy according to the control effect evaluation result to obtain the operation mode switching control strategy.

[0012] In a second aspect, the present application provides an operation mode switching control device for an offshore wind turbine generator set, the operation mode switching control device for an offshore wind turbine generator set comprising:

[0013] The acquisition module is used to collect and preprocess the operating parameters of the target offshore wind turbine in real time to obtain the current status evaluation data and short-term prediction data of the turbine;

[0014] A classification module is used to perform model classification on the historical operation data of the unit acquired in advance to obtain operation mode classification data and a characteristic parameter library;

[0015] An analysis module, used to perform reinforcement learning analysis on the current state evaluation data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode;

[0016] A processing module, used for performing fuzzy control processing on the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence;

[0017] A control module, used to perform multi-objective optimization and coordinated control analysis on the mode switching strategy and the parameter adjustment sequence to obtain precise control signals for each actuator in the target offshore wind turbine;

[0018] The switching module is used to generate an operation mode switching control strategy according to the precise control signals of each actuator, and to perform mode switching control on the target offshore wind turbine group through the mode switching control strategy.

[0019] The third aspect of the present application provides an operation mode switching control device for an offshore wind turbine group, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the operation mode switching control device of the offshore wind turbine group executes the above-mentioned operation mode switching control method for the offshore wind turbine group.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned operation mode switching control method for an offshore wind turbine.

[0021] In the technical solution provided by this application, by real-time collection and preprocessing of the operating parameters of the target offshore wind turbines, combined with algorithms such as wavelet transform denoising, time series analysis, support vector regression and Kalman filtering, an accurate assessment of the current state and short-term trend of the unit is achieved, providing a high-quality data basis for subsequent decision-making. Secondly, the historical operation data is deeply mined and modeled, and algorithms such as principal component analysis, support vector machine, decision tree and fuzzy C-mean are used to construct a comprehensive operation mode classification data and feature parameter library, which greatly improves the accuracy and robustness of pattern recognition. Thirdly, deep neural networks are used for reinforcement learning analysis, and advanced algorithms such as autoencoders, deep Q networks, Monte Carlo tree search and policy gradients are used to achieve intelligent selection of the optimal operation mode in complex environments, significantly improving the adaptability and foresight of decision-making. In addition, the fuzzy control algorithm is introduced to formulate the mode switching strategy, and the smoothness and safety of the switching process are achieved by combining technologies such as adaptive fuzzy PID and fuzzy cognitive maps. In terms of actuator control, multi-objective optimization and collaborative control analysis are adopted, and multi-objective genetic algorithm, fuzzy comprehensive evaluation, model predictive control and H infinite control are used in combination to obtain accurate control signals for each actuator, effectively solve the coupling problem between actuators, and improve the accuracy and stability of control. Finally, when generating the final operation mode switching control strategy, the real-time optimization and dynamic adjustment of the control strategy are realized through the comprehensive application of algorithms such as dynamic programming, fuzzy logic control, sliding mode control and adaptive control, further enhancing the robustness and adaptability of the system. This multi-level and multi-algorithm fusion application not only significantly improves the operating efficiency and reliability of offshore wind turbines under complex sea conditions, but also provides a new technical path for the intelligent operation and maintenance of large-scale offshore wind farms. At the same time, the modular design and adaptive optimization mechanism of this method make it have good scalability and continuous improvement capabilities, and can continuously adapt to new environmental changes and equipment conditions, extend the service life of offshore wind turbines, reduce operation and maintenance costs, and improve power generation efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 This is a schematic diagram of an embodiment of an operation mode switching control method for an offshore wind turbine in an embodiment of the present application;

[0024] Figure 2 Schematic diagram of an embodiment of an operating mode switching control device for an offshore wind turbine in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The embodiment of the present application provides an operation mode switching control method and device for an offshore wind turbine. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the operation mode switching control method of an offshore wind turbine in the embodiment of the present application includes:

[0027] Step S101, real-time collection and preprocessing of operating parameters of the target offshore wind turbine generator set to obtain current state assessment data and short-term prediction data of the turbine generator set;

[0028] Step S102: Model classification is performed on the previously acquired historical operation data of the unit to obtain operation mode classification data and a characteristic parameter library;

[0029] Step S103: Perform reinforcement learning analysis on the current state assessment data and short-term prediction data of the unit through a preset deep neural network to obtain a target operation mode;

[0030] Step S104, performing fuzzy control processing on the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence;

[0031] Step S105, performing multi-objective optimization and coordinated control analysis on the mode switching strategy and parameter adjustment sequence to obtain accurate control signals for each actuator in the target offshore wind turbine;

[0032] Step S106: Generate an operation mode switching control strategy according to the precise control signal of each actuator, and perform mode switching control on the target offshore wind turbine set through the mode switching control strategy.

[0033] It is understandable that the execution subject of the present application may be an operation mode switching control device of an offshore wind turbine, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0034] Specifically, when implementing the operation mode switching control method of an offshore wind turbine, it is first necessary to collect and preprocess the operating parameters of the target offshore wind turbine in real time. This step involves collecting environmental parameters such as wind speed, wind direction, turbulence intensity, wave height, tidal conditions, and unit operating parameters such as power output, blade pitch angle, hub speed, main bearing temperature, gearbox vibration, and generator current. For example, in a 5MW offshore wind turbine, wind speed data may be collected once a second and wave height data may be collected once a minute. These raw data are denoised by the wavelet transform algorithm to eliminate measurement errors and environmental interference. Subsequently, the processed data are short-term predicted by time series analysis and support vector regression algorithm, such as predicting the wind speed change trend in the next 10 minutes. At the same time, the fuzzy inference algorithm is used to evaluate the current operating state of the unit to obtain the current state evaluation result of the unit. These data are further optimized by the Kalman filter algorithm to finally form the unit current state evaluation data and short-term prediction data.

[0035] Model classification is performed on the historical operation data of the units acquired in advance. This step first cleans the historical data to remove outliers and invalid data. Then the data is reduced in dimension and key features are extracted through the principal component analysis algorithm. For example, 10-20 main features may be extracted from hundreds of original parameters. Subsequently, the support vector machine algorithm is used to classify these feature data to obtain preliminary operating mode categories. These categories may include normal power generation mode, maximum power tracking mode, load reduction mode, typhoon protection mode, etc. Rules are extracted from these modes through the decision tree algorithm, and fuzzy boundary processing is performed using the fuzzy C-means algorithm, ultimately forming the operating mode classification data and feature parameter library.

[0036] Reinforcement learning analysis is performed on the unit's current state assessment data and short-term forecast data through a preset deep neural network. This step first uses the autoencoder algorithm to extract and compress the input data, and then maps these features to the state space through the deep Q network algorithm. For example, the unit's operating state may be mapped to a multidimensional space, where each dimension represents a key parameter. Subsequently, the Monte Carlo tree search algorithm is used to explore possible action sequences in this state space, and the policy gradient algorithm is used to continuously optimize the decision-making strategy. Finally, the probability distribution of various possible operating modes is obtained through Bayesian inference, and the optimal target operating mode is selected.

[0037] After the target operation mode is obtained, fuzzy control processing is performed through the preset fuzzy control algorithm. This step first converts the target operation mode into a fuzzy input set, and then performs rule matching and reasoning operations through the fuzzy inference engine. For example, if the target operation mode is "load reduction mode", the fuzzy control algorithm may derive a series of fuzzy control rules based on the current wind speed and power output. These rules are defuzzified to form specific switching strategies and parameter adjustment sequences. At the same time, these parameters are optimized through the adaptive fuzzy PID algorithm to adapt to different sea conditions and unit states.

[0038] Multi-objective optimization and collaborative control analysis are performed on the mode switching strategy and parameter adjustment sequence. This step first solves the Pareto optimal solution set through a multi-objective genetic algorithm, considering multiple objectives such as power generation efficiency, equipment life, and safety. Then the best solution is selected from these optimal solutions through a fuzzy comprehensive evaluation method. For example, in a mode switch, a 10% increase in power generation and a 5% reduction in equipment wear may be considered at the same time. Subsequently, the control signals of each actuator are optimized and robustness enhanced through the model predictive control algorithm and the H-infinity control algorithm to cope with the uncertainty of the offshore environment.

[0039] Finally, the operation mode switching control strategy is generated based on the optimized actuator control signal. This step globally optimizes the control sequence through a dynamic programming algorithm, and then uses a fuzzy logic controller and a sliding mode control algorithm to enhance the robustness of the control strategy. For example, when switching from normal power generation mode to typhoon protection mode, the control strategy may include a series of precise control instructions such as gradually adjusting the blade pitch angle, reducing the generator speed, and adjusting the yaw angle. These instructions are evaluated for safety and adjusted online by the adaptive control algorithm to form the final operation mode switching control strategy, which is used to guide the actual operation control of the target offshore wind turbine.

[0040] In the embodiment of the present application, by real-time acquisition and preprocessing of the operating parameters of the target offshore wind turbine, combined with algorithms such as wavelet transform denoising, time series analysis, support vector regression and Kalman filtering, an accurate assessment of the current state and short-term trend of the unit is achieved, providing a high-quality data basis for subsequent decision-making. Secondly, the historical operation data is deeply mined and modeled, and algorithms such as principal component analysis, support vector machine, decision tree and fuzzy C-mean are used to construct a comprehensive operation mode classification data and feature parameter library, which greatly improves the accuracy and robustness of pattern recognition. Thirdly, a deep neural network is used for reinforcement learning analysis, and through advanced algorithms such as autoencoders, deep Q networks, Monte Carlo tree search and policy gradients, the intelligent selection of the optimal operation mode in a complex environment is achieved, which significantly improves the adaptability and foresight of decision-making. In addition, a fuzzy control algorithm is introduced to formulate a mode switching strategy, and combined with technologies such as adaptive fuzzy PID and fuzzy cognitive maps, the smoothness and safety of the switching process are achieved. In terms of actuator control, multi-objective optimization and collaborative control analysis are adopted, and multi-objective genetic algorithm, fuzzy comprehensive evaluation, model predictive control and H infinite control are used in combination to obtain accurate control signals for each actuator, effectively solve the coupling problem between actuators, and improve the accuracy and stability of control. Finally, when generating the final operation mode switching control strategy, the real-time optimization and dynamic adjustment of the control strategy are realized through the comprehensive application of algorithms such as dynamic programming, fuzzy logic control, sliding mode control and adaptive control, further enhancing the robustness and adaptability of the system. This multi-level and multi-algorithm fusion application not only significantly improves the operating efficiency and reliability of offshore wind turbines under complex sea conditions, but also provides a new technical path for the intelligent operation and maintenance of large-scale offshore wind farms. At the same time, the modular design and adaptive optimization mechanism of this method make it have good scalability and continuous improvement capabilities, and can continuously adapt to new environmental changes and equipment conditions, extend the service life of offshore wind turbines, reduce operation and maintenance costs, and improve power generation efficiency.

[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0042] (1) Real-time monitoring of wind speed, wind direction, turbulence intensity, wave height and tidal conditions of target offshore wind turbines to obtain environmental parameter data;

[0043] (2) Real-time collection of the target offshore wind turbine’s power output, blade pitch angle, hub speed, main bearing temperature, gearbox vibration, and generator current to obtain the unit’s operating parameter data;

[0044] (3) performing denoising processing on the environmental parameter data and the unit operation parameter data to obtain pre-processed environmental parameter data and the unit operation parameter data;

[0045] (4) Perform time series analysis on the preprocessed environmental parameter data to obtain short-term environmental prediction data;

[0046] (5) Statistically analyzing the preprocessed unit operating parameter data to obtain the unit's current operating status data;

[0047] (6) Perform regression analysis on short-term environmental prediction data using support vector regression algorithm to obtain predicted values ​​of environmental parameters;

[0048] (7) Evaluate the current operating status data of the unit through fuzzy reasoning algorithm to obtain the current status evaluation result of the unit;

[0049] (8) Perform data fusion on the predicted values ​​of environmental parameters and the current status assessment results of the unit to obtain comprehensive status prediction data;

[0050] (9) filtering the comprehensive state prediction data through a Kalman filter algorithm to obtain optimized state prediction data;

[0051] (10) Perform threshold judgment on the optimized state prediction data to obtain the unit's current state evaluation data and short-term prediction data.

[0052] Specifically, the environmental parameters and operating parameters of the target offshore wind turbines are comprehensively monitored and collected in real time. Specifically, the wind speed, wind direction, turbulence intensity, wave height and tide conditions are monitored in real time through the anemometer, wind vane, turbulence intensity meter, wave meter and tide gauge installed on the wind turbine, so as to obtain environmental parameter data. For example, in a 5MW offshore wind turbine, wind speed data may be collected once per second, while wave height data may be collected every 10 minutes. At the same time, through the built-in sensor network of the unit, key operating parameters such as power output, blade pitch angle, hub speed, main bearing temperature, gearbox vibration and generator current are collected in real time to obtain the unit operation parameter data. The collection frequency of these parameters may vary, such as power output may be collected once per second, while main bearing temperature may be collected once per minute.

[0053] In order to improve the data quality, the collected environmental parameter data and unit operation parameter data are denoised. This step usually uses the wavelet transform algorithm, which can effectively remove high-frequency noise in the data while retaining the important features of the signal. For example, for wind speed data, the db4 wavelet basis function may be used for 5-layer decomposition to remove the highest frequency detail coefficients, and then reconstruct the denoised wind speed data. In this way, the pre-processed environmental parameter data and unit operation parameter data can be obtained.

[0054] Perform time series analysis on the preprocessed environmental parameter data to obtain short-term environmental forecast data. This step may use the autoregressive integrated moving average model (ARIMA), which can capture the trend and periodic characteristics of the data. For example, for wind speed data, an ARIMA (2,1,2) model may be established to predict the wind speed change trend in the next hour. At the same time, perform statistical analysis on the preprocessed unit operating parameter data, calculate statistics such as mean, standard deviation, kurtosis, etc., and obtain the current operating status data of the unit.

[0055] The short-term environmental forecast data is regressed through the support vector regression (SVR) algorithm to obtain more accurate environmental parameter prediction values. The SVR algorithm is a powerful nonlinear regression tool that can handle high-dimensional data and has good generalization ability. In this case, the SVR model with a radial basis function (RBF) kernel may be used. The input includes parameters such as current wind speed, wind direction, and turbulence intensity, and the output is the wind speed forecast value every 10 minutes in the next hour.

[0056] The current operating status data of the unit is evaluated by the fuzzy inference algorithm to obtain the current status evaluation result of the unit. The fuzzy inference algorithm can handle uncertainty and ambiguity and is suitable for dealing with complex wind turbine status evaluation problems. For example, three fuzzy sets of "low", "medium", and "high" may be defined to describe the power output level, and then the current operating status of the unit is evaluated according to predefined fuzzy rules (such as "if the wind speed is high and the power output is low, the unit efficiency is low").

[0057] The environmental parameter prediction values ​​and the unit current state assessment results are fused to obtain comprehensive state prediction data. This step may use the Dempster-Shafer evidence theory, which can effectively integrate uncertain information from different sources. For example, the uncertainty of wind speed prediction and the uncertainty of unit state assessment may be fused to obtain a more comprehensive state prediction.

[0058] In order to further improve the prediction accuracy, the comprehensive state prediction data is filtered by the Kalman filter algorithm to obtain the optimized state prediction data. Kalman filtering is a recursive state estimation algorithm that can optimally estimate the system state in the presence of measurement noise. In this case, a state space model containing key parameters such as wind speed and power output may be established, and then the Kalman filter algorithm is used to estimate and predict these state variables in real time.

[0059] Finally, the optimized state prediction data is subjected to threshold judgment to obtain the final unit current state assessment data and short-term prediction data. This step may set multiple thresholds, such as wind speed threshold, power curve deviation threshold, etc., to determine whether the unit is in normal operation and whether it may need to switch the operation mode in the short term in the future.

[0060] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0061] (1) Clean the historical operation data of the unit to obtain a valid historical data set;

[0062] (2) Perform time series segmentation on the valid historical data set to obtain multiple time window data;

[0063] (3) Extract features from multiple time window data using the principal component analysis algorithm to obtain dimension-reduced feature data;

[0064] (4) Perform cluster analysis on the reduced-dimensional feature data to obtain preliminary operating mode categories;

[0065] (5) Optimizing the boundaries of the preliminary operating mode categories through a support vector machine algorithm to obtain optimized operating mode categories;

[0066] (6) labeling the optimized operation mode categories to obtain labeled operation mode data;

[0067] (7) extracting rules from the marked operation mode data through a decision tree algorithm to obtain operation mode classification rules;

[0068] (8) Cross-validating the operation mode classification rules to obtain validated classification rules;

[0069] (9) The verified classification rules are fuzzy-processed by the fuzzy C-means algorithm to obtain fuzzy classification rules;

[0070] (10) Perform correlation analysis on the fuzzy classification rules and dimensionality reduction feature data to obtain the operation mode classification data and feature parameter library.

[0071] Specifically, in the operation mode switching control method of offshore wind turbines, the processing and analysis of historical operation data is a crucial step. First, the historical operation data of the unit is cleaned, which includes removing outliers, processing missing data, and eliminating duplicate records. For example, for a 5MW offshore wind turbine that has been in operation for 5 years, there may be millions of historical records. During the data cleaning process, records with wind speeds exceeding 50m / s will be marked as outliers and deleted, and missing power output data may be filled in by linear interpolation methods. In this way, a reliable and valid historical data set is obtained.

[0072] The valid historical data set is segmented into time series to obtain multiple time window data. This step is intended to capture the operating mode characteristics at different time scales. For example, three time windows of 1 hour, 1 day, and 1 week may be selected for segmentation. For each time window, the corresponding operating parameter data is extracted, such as average wind speed, average power output, maximum blade pitch angle, etc.

[0073] The principal component analysis (PCA) algorithm is used to extract features from multiple time window data to obtain dimension-reduced feature data. The PCA algorithm can project high-dimensional data into a low-dimensional space and retain the most important feature information. In this case, 10 principal components may be extracted from the original 50 or more parameters, which can explain more than 90% of the variance of the original data. In this way, not only the dimension of the data is reduced, but also the most critical features are highlighted.

[0074] Perform cluster analysis on the feature data after dimensionality reduction to obtain preliminary operating mode categories. This step may use the K-means clustering algorithm to cluster similar operating states together by calculating the Euclidean distance between data points. For example, five preliminary operating mode categories may be obtained, including normal power generation mode, maximum power tracking mode, load reduction mode, standby mode, and fault mode.

[0075] The support vector machine (SVM) algorithm is used to optimize the boundaries of the preliminary operating mode categories to obtain the optimized operating mode categories. The SVM algorithm can construct the optimal classification hyperplane in high-dimensional space and is particularly suitable for dealing with nonlinear classification problems. In this case, the radial basis function (RBF) kernel SVM may be used to optimize the kernel function parameters and penalty coefficients through the grid search method to obtain a more accurate category boundary.

[0076] The optimized operating mode categories are labeled to obtain labeled operating mode data. This step assigns a unique identifier to each operating mode for subsequent analysis and processing. For example, the normal power generation mode may be labeled as "1", the maximum power tracking mode is labeled as "2", and so on.

[0077] The decision tree algorithm is used to extract rules from the marked operation mode data to obtain the operation mode classification rules. The decision tree algorithm can generate intuitive and easy-to-interpret classification rules. In this example, the C4.5 decision tree algorithm may be used to generate a series of "if-then" rules, such as "If the average wind speed is <3m / s and the power output is <100kW, then it is in standby mode."

[0078] Cross-validate the operation mode classification rules to obtain the verified classification rules. This step usually uses the 10-fold cross-validation method, which divides the data set into 10 parts, and uses 9 parts as training sets and 1 part as test sets in turn to evaluate the generalization ability of the classification rules. In this way, the classification rules with the highest accuracy and the most stability can be screened out.

[0079] Then, the verified classification rules are fuzzy-processed by the fuzzy C-means (FCM) algorithm to obtain fuzzy classification rules. The FCM algorithm allows data points to belong to multiple categories to varying degrees. This soft classification method is more suitable for handling the complex operating states of offshore wind turbines. For example, an operating state may have 80% of the normal power generation mode characteristics and 20% of the load reduction mode characteristics at the same time.

[0080] Finally, the fuzzy classification rules and the dimension reduction feature data are analyzed for association, and the operation mode classification data and feature parameter library are obtained. This step may use the Apriori algorithm to mine the association rules between feature parameters and operation modes. For example, it may be found that the set of features "high wind speed + high power output + medium blade pitch angle" is highly correlated with the maximum power tracking mode.

[0081] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0082] (1) Fusing the unit's current status assessment data and short-term prediction data to obtain comprehensive status data;

[0083] (2) extracting features from the comprehensive state data through an autoencoder algorithm to obtain a compressed feature vector;

[0084] (3) Normalizing the compressed feature vector to obtain standardized feature data;

[0085] (4) Perform state space mapping on the standardized feature data through a deep Q-network algorithm to obtain a state representation;

[0086] (5) Discretize the action space of the state representation to obtain a set of candidate actions;

[0087] (6) Explore the candidate action set through the Monte Carlo tree search algorithm to obtain the optimal action sequence;

[0088] (7) Evaluate the reward function of the optimal action sequence to obtain the action value estimate;

[0089] (8) The action value estimate is updated by the policy gradient algorithm to obtain the optimized policy parameters;

[0090] (9) Perform Bayesian inference on the optimized strategy parameters to obtain the probability distribution of the operating mode;

[0091] (10) The maximum a posteriori probability criterion is used to make a decision on the probability distribution of the operating mode and obtain the target operating mode.

[0092] Specifically, in the operation mode switching control method of offshore wind turbines, deep learning analysis of the current state evaluation data and short-term prediction data of the unit is a key step. First, the current state evaluation data of the unit (such as current wind speed, power output, blade pitch angle, etc.) and short-term prediction data (such as wind speed forecast and power forecast for the next hour) are fused to obtain comprehensive state data. This step uses the weighted average method. For example, for a 5MW offshore wind turbine, the current wind speed and the predicted wind speed after 1 hour are assigned weights of 0.7 and 0.3 respectively, forming a more comprehensive state description.

[0093] The autoencoder algorithm is used to extract features from the comprehensive state data to obtain a compressed feature vector. The autoencoder is an unsupervised learning algorithm that learns the effective representation of data by compressing the input data and then reconstructing it. In this example, a three-layer autoencoder is used, with 50 nodes in the input layer (corresponding to 50 state parameters), 20 nodes in the hidden layer, and 50 nodes reconstructed back to the output layer. Through training, the autoencoder can compress the 50-dimensional input data into a 20-dimensional feature vector, retaining the main feature information of the data.

[0094] Normalize the compressed feature vector to obtain standardized feature data. This step uses the min-max normalization method to map each feature value to the interval [0,1]. For example, for the wind speed feature, the minimum value 0m / s is mapped to 0, the maximum value 30m / s is mapped to 1, and other values ​​are mapped proportionally. This standardization process helps to eliminate the dimensional differences between different features and improve the effect of subsequent learning algorithms.

[0095] The state representation is obtained by mapping the standardized feature data into state space through the Deep Q Network (DQN) algorithm. DQN is an algorithm in reinforcement learning that combines Q learning and deep neural networks. In this example, the DQN network contains 3 fully connected layers, each with 64, 32 and 16 neurons respectively. The input is 20-dimensional standardized feature data, and the output is the Q value of different actions. Through this network, the continuous state space is mapped to the discrete action value space.

[0096] The action space of the state representation is discretized to obtain a candidate action set. For example, the operation modes of wind turbines are divided into five types: normal power generation, maximum power tracking, load reduction, standby and emergency shutdown. Each mode is further subdivided into multiple control actions, such as adjusting the blade pitch angle, changing the generator speed, etc., and finally a candidate action set containing 20 discrete actions is formed.

[0097] The Monte Carlo Tree Search (MCTS) algorithm is used to explore the candidate action set and obtain the optimal action sequence. MCTS is a sampling-based search algorithm that is particularly suitable for large-scale decision-making problems. In this example, starting from the current state, different action sequences are simulated 1000 times, each simulation is 10 steps deep, and the optimal action sequence is selected based on the cumulative reward.

[0098] The reward function is used to evaluate the optimal action sequence to obtain an estimate of the action value. The reward function takes into account factors such as power generation, equipment life, and safety. For example, 1 point is awarded for increasing power generation by 1 kWh, 2 points are awarded for reducing equipment wear by 0.1%, and 10 points are awarded for avoiding an emergency shutdown. In this way, the cumulative reward for each action sequence is calculated as its value estimate.

[0099] The policy gradient algorithm is used to perform gradient updates on the action value estimates to obtain the optimized policy parameters. The policy gradient algorithm directly optimizes the policy function rather than indirectly optimizing it through the value function. In this example, the REINFORCE algorithm is used to calculate the policy gradient based on the cumulative reward of each action sequence, and then update the parameters of the policy network. This process is repeated 1000 times to continuously optimize the policy function.

[0100] The optimized strategy parameters are inferred by Bayesian method to obtain the probability distribution of the operation mode. Bayesian inference can quantify the uncertainty of decision making. In this case, the variational inference method is used to approximate the posterior distribution. For example, for the current state, the following probability distribution may be obtained: normal power generation mode 0.7, maximum power tracking mode 0.2, load reduction mode 0.1.

[0101] Finally, the maximum a posteriori probability criterion is used to make a decision on the probability distribution of the operating mode and obtain the target operating mode. Simply select the mode with the highest probability as the target operating mode. In the above example, the normal power generation mode will be selected as the target operating mode.

[0102] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0103] (1) Extract features of the target operation mode and obtain a mode feature vector;

[0104] (2) Fuzzify the pattern feature vector to obtain a fuzzy input set;

[0105] (3) The fuzzy input set is matched by the fuzzy inference engine to obtain the activation rule set;

[0106] (4) Perform fuzzy reasoning operations on the activation rule set to obtain a fuzzy output set;

[0107] (5) Defuzzify the fuzzy output set using the centroid method to obtain a preliminary switching strategy;

[0108] (6) Conduct a safety assessment on the preliminary switching strategy and obtain a safe switching range;

[0109] (7) Optimizing the parameters of the safety switching interval through an adaptive fuzzy PID algorithm to obtain optimized switching parameters;

[0110] (8) arranging the optimized switching parameters in time sequence to obtain a parameter adjustment sequence;

[0111] (9) Perform causal analysis on the parameter adjustment sequence through the fuzzy cognitive graph algorithm to obtain the switching influencing factor;

[0112] (10) The switching influencing factors and parameter adjustment sequences are comprehensively processed to obtain the mode switching strategy and parameter adjustment sequence.

[0113] Specifically, the target operation mode is feature extracted to obtain a mode feature vector. For example, for a 5MW offshore wind turbine in maximum power tracking mode, the feature vector may include key parameters such as wind speed, power output, blade pitch angle, hub speed, etc., and the specific values ​​may be [12m / s, 4.5MW, 5°, 15rpm].

[0114] The pattern feature vector is fuzzified to obtain a fuzzy input set. This step converts the exact value into a fuzzy set. Taking wind speed as an example, three fuzzy sets are defined: "low", "medium", and "high", using a triangular membership function. For a wind speed of 12m / s, the membership may be [0, 0.6, 0.4], indicating that its membership in the "medium" fuzzy set is 0.6, and its membership in the "high" fuzzy set is 0.4.

[0115] Then, the fuzzy input set is matched by the fuzzy inference engine to obtain the activated rule set. The fuzzy inference engine contains predefined IF-THEN rules, such as "IF wind speed is medium AND power output is high THEN maintain the current blade pitch angle". According to the input fuzzy set, all activated rules are selected to form the activated rule set.

[0116] Perform fuzzy inference operations on the activated rule set to obtain the fuzzy output set. This step uses the Mamdani inference method to synthesize the outputs of each activated rule through the minimum-maximum synthesis operation. For example, if there are two rules that both affect the adjustment of the blade pitch angle, then take the union of the outputs of these two rules as the final fuzzy output set.

[0117] The fuzzy output set is defuzzified by the centroid method to obtain the preliminary switching strategy. The centroid method calculates the centroid of the fuzzy output set and uses it as the precise output value. For example, for the adjustment of the blade pitch angle, an accurate value of 2° may be obtained, indicating that the pitch angle needs to be increased by 2°.

[0118] Perform a safety assessment on the preliminary switching strategy to obtain a safe switching range. This step takes into account the physical limitations and safety margins of the unit. For example, the safe adjustment range of the blade pitch angle may be [-5°, +5°], so a 2° adjustment is within the safe range.

[0119] The adaptive fuzzy PID algorithm is used to optimize the parameters of the safe switching interval to obtain the optimized switching parameters. The adaptive fuzzy PID algorithm combines the advantages of fuzzy control and PID control, and can dynamically adjust the PID parameters according to the current error and the error change rate. For example, when the error is large, the proportional coefficient Kp is increased; when the error changes rapidly, the differential coefficient Kd is increased. In this way, a more accurate blade pitch angle adjustment value, such as 2.3°, is obtained.

[0120] The optimized switching parameters are arranged in time sequence to obtain a parameter adjustment sequence. This step takes into account the time sequence and interdependence of parameter adjustment. For example, the blade pitch angle is adjusted first, then the generator speed is adjusted, and finally the converter parameters are adjusted to form an orderly parameter adjustment sequence.

[0121] The fuzzy cognitive graph algorithm is used to analyze the causal relationship of the parameter adjustment sequence and obtain the switching influence factor. A fuzzy cognitive graph is a directed graph used to describe the causal relationship between concepts. In this step, a fuzzy cognitive graph containing nodes such as wind speed, power output, and blade pitch angle is constructed, and the degree of influence of each parameter adjustment on the system state is obtained through iterative calculation. For example, it may be found that the influence factor of the blade pitch angle adjustment on the power output is 0.8, while the influence factor on the equipment load is 0.6.

[0122] Finally, the switching influencing factors and parameter adjustment sequence are comprehensively processed to obtain the final mode switching strategy and parameter adjustment sequence. This step uses the influencing factors as weights to perform weighted optimization on the parameter adjustment sequence. For example, considering the high impact of blade pitch angle adjustment on power output, the adjustment range may be slightly reduced to 2.1° to achieve a balance between improving power generation efficiency and reducing equipment load.

[0123] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0124] (1) Decouple the mode switching strategy and parameter adjustment sequence to obtain the initial control target of each actuator;

[0125] (2) Setting constraints on the initial control objectives of each actuator to obtain the boundary conditions of the optimization problem;

[0126] (3) Solve the optimization problem through a multi-objective genetic algorithm and obtain the Pareto optimal solution set;

[0127] (4) Perform hierarchical analysis on the Pareto optimal solution set to obtain the weight coefficient of each objective;

[0128] (5) Evaluate the Pareto optimal solution set through fuzzy comprehensive evaluation method and obtain the optimal solution;

[0129] (6) Allocate actuators to the optimal solution and obtain preliminary control signals for each actuator;

[0130] (7) feedforward optimization of the preliminary control signals of each actuator is performed through a model predictive control algorithm to obtain an optimized control sequence;

[0131] (8) Performing stability analysis on the optimized control sequence to obtain stability evaluation results;

[0132] (9) The robustness of the optimized control sequence is enhanced by the H-infinity control algorithm to obtain a robust control signal;

[0133] (10) Dynamically compensate the robust control signal to obtain the precise control signal of each actuator.

[0134] Specifically, in the operation mode switching control method of offshore wind turbines, multi-objective optimization and coordinated control analysis of the mode switching strategy and parameter adjustment sequence are key steps to ensure efficient and accurate control. First, the mode switching strategy and parameter adjustment sequence are decoupled and analyzed to obtain the initial control target of each actuator. For example, for a 5MW offshore wind turbine to switch from normal power generation mode to maximum power tracking mode, three main actuators may be involved: the blade pitch system, the generator speed control system, and the converter control system. After decoupling analysis, the blade pitch angle adjustment target is increased by 2.1°, the generator speed is increased by 5rpm, and the converter power factor is adjusted to 0.95.

[0135] Next, constraints are set for the initial control objectives of each actuator to obtain the boundary conditions of the optimization problem. This step takes into account the physical limitations and safety boundaries of the actuator. For example, the blade pitch angle adjustment rate does not exceed 2° / s, the generator speed change rate does not exceed 10rpm / s, and the converter power factor adjustment range is 0.9 to 1.0. These constraints constitute the boundaries of the optimization problem.

[0136] Then, the optimization problem is solved by a multi-objective genetic algorithm to obtain the Pareto optimal solution set. The multi-objective genetic algorithm is an evolutionary algorithm that can optimize multiple objective functions at the same time. In this example, three objective functions are set: maximizing power generation, minimizing mechanical load, and minimizing control action amplitude. The algorithm simulates the natural selection process and evolves for 100 generations, each generation containing 200 individuals, and finally obtains a set of non-dominated solutions, namely the Pareto optimal solution set.

[0137] Perform a hierarchical analysis on the Pareto optimal solution set to obtain the weight coefficient of each objective. The hierarchical analysis method constructs a judgment matrix and calculates the eigenvector to obtain the relative importance of each objective. For example, the weights of maximizing power generation, minimizing mechanical load, and minimizing control action amplitude may be 0.5, 0.3, and 0.2, respectively.

[0138] The Pareto optimal solution set is evaluated by the fuzzy comprehensive evaluation method to obtain the optimal solution. The fuzzy comprehensive evaluation method comprehensively considers multiple evaluation factors and gives a comprehensive evaluation result. In this step, each solution in the Pareto optimal solution set is weighted and scored according to the weight coefficient obtained previously, and the solution with the highest score is selected as the optimal solution.

[0139] The optimal solution is assigned to the actuators to obtain the preliminary control signals of each actuator. For example, the optimal solution may require the blade pitch angle to increase by 2.0°, the generator speed to increase by 4rpm, and the converter power factor to be adjusted to 0.96. These specific values ​​constitute the preliminary control signals of each actuator.

[0140] The preliminary control signals of each actuator are feedforward optimized through the model predictive control (MPC) algorithm to obtain the optimized control sequence. The MPC algorithm predicts future outputs based on the system model and optimizes the control sequence in the rolling time domain. In this example, a state space model including wind speed, blade pitch angle, generator speed and power output is established. The prediction time domain is set to 10s and the control time domain is set to 5s. The optimized control sequence is obtained by solving the quadratic programming problem.

[0141] Perform stability analysis on the optimized control sequence to obtain stability evaluation results. This step evaluates the stability of the system after the control sequence is executed by calculating the eigenvalue or Lyapunov function of the system. If it is found that some control actions may cause system instability, corresponding adjustments need to be made.

[0142] The robustness of the optimized control sequence is enhanced by the H-infinity control algorithm to obtain a robust control signal. The H-infinity control algorithm aims to design a controller so that the closed-loop system can maintain stability and performance under the worst-case scenario. In this example, an H-infinity controller is designed to enhance the robustness of the control sequence by considering uncertainties such as wind speed measurement errors and actuator response delays.

[0143] Finally, the robust control signal is dynamically compensated to obtain the precise control signal for each actuator. Dynamic compensation takes into account the dynamic characteristics of the actuator and the nonlinearity of the system, and further improves the control accuracy by combining feedforward compensation and feedback correction. For example, considering the mechanical inertia of the blade pitch system, it may be necessary to add an advance amount to the control signal to make the actual pitch angle change closer to the target value.

[0144] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0145] (1) Arrange the precise control signals of each actuator in time sequence to obtain a control sequence;

[0146] (2) Optimize the control sequence through dynamic programming algorithm to obtain the optimal control path;

[0147] (3) Segment the optimal control path to obtain the stage control target;

[0148] (4) Perform fuzzy reasoning on the stage control target through the fuzzy logic controller to obtain the fuzzy control rules;

[0149] (5) Defuzzify the fuzzy control rules to obtain accurate control instructions;

[0150] (6) The robustness of the precise control command is enhanced through the sliding mode control algorithm to obtain an anti-disturbance control strategy;

[0151] (7) Conduct safety assessment on the disturbance control strategy and obtain the safety control boundary;

[0152] (8) Online adjustment of control parameters within the safety control boundary is performed through an adaptive control algorithm to obtain an adaptive control strategy;

[0153] (9) Conduct simulation verification on the adaptive control strategy and obtain the control effect evaluation results;

[0154] (10) The adaptive control strategy is fine-tuned according to the control effect evaluation results to obtain the operation mode switching control strategy.

[0155] Specifically, in the operation mode switching control method of offshore wind turbines, the last key step is to generate and optimize the operation mode switching control strategy. First, the precise control signals of each actuator are arranged in time sequence to obtain a control sequence. For example, for a 5MW offshore wind turbine switching from normal power generation mode to maximum power tracking mode, the control sequence may include: the blade pitch angle starts to adjust at t = 0s, the generator speed starts to change at t = 2s, and the converter parameters start to adjust at t = 4s, and each action lasts for 10s.

[0156] Next, the control sequence is optimized by a dynamic programming algorithm to obtain the optimal control path. The dynamic programming algorithm solves the optimization problem by breaking down a complex problem into a series of sub-problems. In this example, the 10-second control process is divided into 50 time steps, each time step is 0.2 seconds. For each time step, all possible combinations of control actions are considered, and the cumulative reward (such as the weighted sum of the increase in power generation and the reduction in mechanical stress) is calculated, and finally an optimal control path is obtained.

[0157] Then, the optimal control path is segmented to obtain the stage control target. For example, the entire control process may be divided into three stages: initial adjustment stage (0-3s), transition stage (3-7s) and stable stage (7-10s). Each stage has its specific control target, such as rapid adjustment of blade pitch angle in the initial stage, synchronous adjustment of generator speed and converter parameters in the transition stage, and fine-tuning of various parameters in the stable stage.

[0158] The fuzzy logic controller performs fuzzy reasoning on the stage control target and obtains the fuzzy control rules. The fuzzy logic controller uses linguistic variables and fuzzy rules to describe the control strategy. For example, for the blade pitch angle adjustment, there may be a rule like this: "If the wind speed increases quickly and the power deviation is large, then increase the pitch angle quickly." This fuzzy rule can better handle the nonlinearity and uncertainty of the system.

[0159] Defuzzification of fuzzy control rules can be used to obtain precise control instructions. The defuzzification process converts fuzzy outputs into specific values. A common method is the centroid method, which calculates the centroid of the fuzzy output set. For example, for blade pitch angle adjustment, defuzzification may result in a specific adjustment amount, such as an increase of 2.3°.

[0160] The robustness of the precise control instructions is enhanced by the sliding mode control algorithm to obtain an anti-disturbance control strategy. Sliding mode control is a nonlinear control method that can effectively cope with system parameter changes and external disturbances. In this example, considering the complexity of offshore wind conditions, sliding mode control is used to enhance the anti-disturbance capability of the blade pitch system and the generator speed control system to ensure stable control when the wind speed changes suddenly.

[0161] The anti-disturbance control strategy is evaluated for safety and the safety control boundary is obtained. This step takes into account the physical limitations and safety requirements of the equipment. For example, the blade pitch angle adjustment speed cannot exceed 3° / s, and the rate of change of the generator speed cannot exceed 20rpm / s. Through simulation analysis, the safety control boundary under these constraints is determined.

[0162] The adaptive control algorithm is used to adjust the control parameters within the safety control boundary online to obtain an adaptive control strategy. Adaptive control can adjust the control parameters in real time according to the system response. In this example, the model reference adaptive control (MRAC) method may be used to dynamically adjust the controller gain parameters to adapt to different operating conditions according to the deviation of the actual wind speed and power output.

[0163] The adaptive control strategy is simulated and verified to obtain the control effect evaluation results. A high-fidelity wind turbine simulation model is used to simulate different wind and sea conditions to verify the effectiveness of the control strategy. Evaluation indicators include power tracking accuracy, mechanical load, control action smoothness, etc. For example, in a 10-minute simulation, it may be found that the new control strategy increases power generation by 2.5% compared to the traditional method, while reducing blade load fluctuations by 5%.

[0164] Finally, the adaptive control strategy is fine-tuned based on the control effect evaluation results to obtain the final operation mode switching control strategy. This step may involve adjusting the controller parameters, such as the proportional, integral, and differential coefficients of the PID controller, or the sliding surface parameters of the sliding mode controller. For example, if the control response is found to be too aggressive, the proportional coefficient may be slightly reduced and the integral time constant may be increased to obtain a smoother control effect.

[0165] The above describes the operation mode switching control method of the offshore wind turbine in the embodiment of the present application. The following describes the operation mode switching control device of the offshore wind turbine in the embodiment of the present application. Figure 2In the embodiment of the present application, an embodiment of the operation mode switching control device of the offshore wind turbine generator set includes:

[0166] The acquisition module 201 is used to collect and pre-process the operating parameters of the target offshore wind turbine in real time to obtain the current state evaluation data and short-term prediction data of the turbine;

[0167] The classification module 202 is used to perform model classification on the pre-acquired historical operation data of the unit to obtain operation mode classification data and a characteristic parameter library;

[0168] The analysis module 203 is used to perform reinforcement learning analysis on the current state evaluation data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode;

[0169] The processing module 204 is used to perform fuzzy control processing on the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence;

[0170] A control module 205 is used to perform multi-objective optimization and coordinated control analysis on the mode switching strategy and the parameter adjustment sequence to obtain accurate control signals for each actuator in the target offshore wind turbine;

[0171] The switching module 206 is used to generate an operation mode switching control strategy according to the precise control signal of each actuator, and perform mode switching control on the target offshore wind turbine group through the mode switching control strategy.

[0172] Through the collaboration of the above components, the operating parameters of the target offshore wind turbines are collected and preprocessed in real time, combined with algorithms such as wavelet transform denoising, time series analysis, support vector regression and Kalman filtering, to achieve an accurate assessment of the current state and short-term trend of the unit, providing a high-quality data basis for subsequent decision-making. Secondly, the historical operating data is deeply mined and modeled, and algorithms such as principal component analysis, support vector machine, decision tree and fuzzy C-mean are used to build a comprehensive operating mode classification data and feature parameter library, which greatly improves the accuracy and robustness of pattern recognition. Thirdly, deep neural networks are used for reinforcement learning analysis, and advanced algorithms such as autoencoders, deep Q networks, Monte Carlo tree search and policy gradients are used to achieve intelligent selection of the optimal operating mode in complex environments, significantly improving the adaptability and foresight of decision-making. In addition, fuzzy control algorithms are introduced to formulate mode switching strategies, and technologies such as adaptive fuzzy PID and fuzzy cognitive maps are combined to achieve smoothness and safety of the switching process. In terms of actuator control, multi-objective optimization and collaborative control analysis are adopted, and multi-objective genetic algorithm, fuzzy comprehensive evaluation, model predictive control and H infinite control are used in combination to obtain accurate control signals for each actuator, effectively solve the coupling problem between actuators, and improve the accuracy and stability of control. Finally, when generating the final operation mode switching control strategy, the real-time optimization and dynamic adjustment of the control strategy are realized through the comprehensive application of algorithms such as dynamic programming, fuzzy logic control, sliding mode control and adaptive control, further enhancing the robustness and adaptability of the system. This multi-level and multi-algorithm fusion application not only significantly improves the operating efficiency and reliability of offshore wind turbines under complex sea conditions, but also provides a new technical path for the intelligent operation and maintenance of large-scale offshore wind farms. At the same time, the modular design and adaptive optimization mechanism of this method make it have good scalability and continuous improvement capabilities, and can continuously adapt to new environmental changes and equipment conditions, extend the service life of offshore wind turbines, reduce operation and maintenance costs, and improve power generation efficiency.

[0173] The present application also provides an operation mode switching control device for an offshore wind turbine group, the operation mode switching control device for the offshore wind turbine group comprising a memory and a processor, the memory storing computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the operation mode switching control method for the offshore wind turbine group in the above-mentioned embodiments.

[0174] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein instructions are stored in the computer, and when the instructions are executed on a computer, the computer executes the steps of the operating mode switching control method of the offshore wind turbine.

[0175] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0176] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0177] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for controlling the switching of operating modes of an offshore wind turbine generator system, characterized in that: The operation mode switching control method of the offshore wind turbine generator system comprises: Real-time collection and preprocessing of the operating parameters of the target offshore wind turbines to obtain the current status assessment data and short-term forecast data of the units; Perform model classification on the historical operation data of the unit acquired in advance to obtain the operation mode classification data and characteristic parameter library; Performing reinforcement learning analysis on the current state assessment data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode; Performing fuzzy control processing on the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence; Performing multi-objective optimization and coordinated control analysis on the mode switching strategy and the parameter adjustment sequence to obtain precise control signals for each actuator in the target offshore wind turbine; An operation mode switching control strategy is generated according to the precise control signals of each actuator, and the target offshore wind turbine is subjected to mode switching control through the mode switching control strategy.

2. The method for controlling the switching of the operation mode of an offshore wind turbine according to claim 1, characterized in that: The real-time collection and preprocessing of the operating parameters of the target offshore wind turbine unit to obtain the unit's current state assessment data and short-term prediction data include: Real-time monitoring of wind speed, wind direction, turbulence intensity, wave height and tidal conditions of the target offshore wind turbines to obtain environmental parameter data; The power output, blade pitch angle, hub speed, main bearing temperature, gearbox vibration and generator current of the target offshore wind turbine are collected in real time to obtain the unit operation parameter data; Performing denoising processing on the environmental parameter data and the unit operation parameter data to obtain pre-processed environmental parameter data and unit operation parameter data; Performing time series analysis on the preprocessed environmental parameter data to obtain short-term environmental prediction data; Performing statistical analysis on the preprocessed unit operating parameter data to obtain the unit's current operating status data; Performing regression analysis on the short-term environmental prediction data by using a support vector regression algorithm to obtain environmental parameter prediction values; The current operating state data of the unit is evaluated by a fuzzy inference algorithm to obtain a current state evaluation result of the unit; Performing data fusion on the environmental parameter prediction value and the current state assessment result of the unit to obtain comprehensive state prediction data; Performing filtering processing on the comprehensive state prediction data by using a Kalman filter algorithm to obtain optimized state prediction data; A threshold value judgment is performed on the optimized state prediction data to obtain the current state evaluation data of the unit and the short-term prediction data.

3. The operation mode switching control method of an offshore wind turbine according to claim 1, characterized in that: The previously acquired historical operation data of the unit is subjected to model classification to obtain operation mode classification data and a characteristic parameter library, including: Performing data cleaning on the historical operation data of the unit to obtain a valid historical data set; Performing time series segmentation on the valid historical data set to obtain multiple time window data; Extracting features from the multiple time window data using a principal component analysis algorithm to obtain dimension-reduced feature data; Performing cluster analysis on the dimension-reduced feature data to obtain preliminary operation mode categories; Performing boundary optimization on the preliminary operation mode category by using a support vector machine algorithm to obtain an optimized operation mode category; Labeling the optimized operation mode category to obtain labeled operation mode data; Extracting rules from the marked operation mode data by using a decision tree algorithm to obtain operation mode classification rules; Cross-validating the operation mode classification rules to obtain validated classification rules; Performing fuzzy boundary processing on the verified classification rules by using a fuzzy C-means algorithm to obtain fuzzy classification rules; The fuzzy classification rules and the dimension reduction feature data are subjected to association analysis to obtain the operation mode classification data and the feature parameter library.

4. The method for controlling the switching of the operation mode of an offshore wind turbine according to claim 1, characterized in that: The method of performing reinforcement learning analysis on the current state assessment data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode includes: Performing data fusion on the current status assessment data of the unit and the short-term prediction data to obtain comprehensive status data; Extracting features from the comprehensive state data using an autoencoder algorithm to obtain a compressed feature vector; Normalizing the compressed feature vector to obtain standardized feature data; Performing state space mapping on the standardized feature data by a deep Q network algorithm to obtain a state representation; Discretizing the state representation into an action space to obtain a candidate action set; The candidate action set is explored by using a Monte Carlo tree search algorithm to obtain an optimal action sequence; Performing a reward function evaluation on the optimal action sequence to obtain an action value estimate; Performing gradient update on the action value estimate by using a policy gradient algorithm to obtain optimized policy parameters; Performing Bayesian inference on the optimized strategy parameters to obtain a probability distribution of the operation mode; The operation mode probability distribution is decided by the maximum a posteriori probability criterion to obtain the target operation mode.

5. The method for controlling the switching of the operation mode of an offshore wind turbine according to claim 1, characterized in that: The fuzzy control processing of the target operation mode is performed by a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence, including: Extracting features of the target operation mode to obtain a mode feature vector; Performing fuzzification processing on the pattern feature vector to obtain a fuzzy input set; Performing rule matching on the fuzzy input set through a fuzzy inference engine to obtain an activation rule set; Performing fuzzy reasoning operation on the activation rule set to obtain a fuzzy output set; Defuzzifying the fuzzy output set by using the centroid method to obtain a preliminary switching strategy; Performing a safety assessment on the preliminary switching strategy to obtain a safe switching interval; Optimizing the parameters of the safety switching interval by using an adaptive fuzzy PID algorithm to obtain optimized switching parameters; Arranging the optimized switching parameters in time sequence to obtain a parameter adjustment sequence; Performing a causal analysis on the parameter adjustment sequence by using a fuzzy cognitive graph algorithm to obtain a switching influencing factor; The switching influencing factors and the parameter adjustment sequence are comprehensively processed to obtain the mode switching strategy and the parameter adjustment sequence.

6. The method for controlling the switching of the operation mode of an offshore wind turbine according to claim 1, characterized in that: The multi-objective optimization and coordinated control analysis of the mode switching strategy and the parameter adjustment sequence to obtain accurate control signals for each actuator in the target offshore wind turbine unit includes: Decoupling analysis is performed on the mode switching strategy and the parameter adjustment sequence to obtain an initial control target for each actuator; Setting constraint conditions for the initial control targets of the actuators to obtain boundary conditions for the optimization problem; Solving the optimization problem by using a multi-objective genetic algorithm to obtain a Pareto optimal solution set; Performing hierarchical analysis on the Pareto optimal solution set to obtain the weight coefficient of each objective; Evaluate the Pareto optimal solution set by a fuzzy comprehensive evaluation method to obtain the optimal solution; Allocating actuators to the optimal solution to obtain preliminary control signals for each actuator; Performing feedforward optimization on the preliminary control signals of each actuator by using a model predictive control algorithm to obtain an optimized control sequence; Performing stability analysis on the optimized control sequence to obtain a stability evaluation result; The optimized control sequence is robustly enhanced by an H-infinity control algorithm to obtain a robust control signal; Dynamically compensate the robust control signal to obtain accurate control signals for each actuator.

7. The method for controlling the switching of the operation mode of an offshore wind turbine according to claim 1, characterized in that: The generating of the operation mode switching control strategy according to the precise control signal of each actuator, and performing mode switching control on the target offshore wind turbine set by means of the mode switching control strategy, comprises: Arranging the precise control signals of the actuators in time sequence to obtain a control sequence; Optimizing the control sequence by a dynamic programming algorithm to obtain an optimal control path; The optimal control path is processed in sections to obtain a stage control target; Performing fuzzy reasoning on the control target of the stage through a fuzzy logic controller to obtain a fuzzy control rule; Defuzzifying the fuzzy control rules to obtain precise control instructions; The precise control instruction is robustly enhanced by a sliding mode control algorithm to obtain an anti-disturbance control strategy; Performing a safety assessment on the anti-disturbance control strategy to obtain a safety control boundary; The control parameters within the safety control boundary are adjusted online by an adaptive control algorithm to obtain an adaptive control strategy; Performing simulation verification on the adaptive control strategy to obtain a control effect evaluation result; The adaptive control strategy is fine-tuned according to the control effect evaluation result to obtain the operation mode switching control strategy.

8. An operating mode switching control device for an offshore wind turbine generator set, characterized in that: The operation mode switching control device of the offshore wind turbine generator set comprises: The acquisition module is used to collect and preprocess the operating parameters of the target offshore wind turbine in real time to obtain the current status evaluation data and short-term prediction data of the turbine; A classification module is used to perform model classification on the historical operation data of the unit acquired in advance to obtain operation mode classification data and a characteristic parameter library; An analysis module, used to perform reinforcement learning analysis on the current state evaluation data of the unit and the short-term prediction data through a preset deep neural network to obtain a target operation mode; A processing module, used for performing fuzzy control processing on the target operation mode through a preset fuzzy control algorithm to obtain a mode switching strategy and a parameter adjustment sequence; A control module, used to perform multi-objective optimization and coordinated control analysis on the mode switching strategy and the parameter adjustment sequence to obtain precise control signals for each actuator in the target offshore wind turbine; The switching module is used to generate an operation mode switching control strategy according to the precise control signals of each actuator, and to perform mode switching control on the target offshore wind turbine group through the mode switching control strategy.

9. An operation mode switching control device for an offshore wind turbine generator set, characterized in that: The operation mode switching control device of the offshore wind turbine generator system comprises: a memory and at least one processor, wherein the memory stores instructions; The at least one processor calls the instruction in the memory so that the operation mode switching control device of the offshore wind turbine group executes the operation mode switching control method of the offshore wind turbine group according to any one of claims 1-7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, an operation mode switching control method for an offshore wind turbine set according to any one of claims 1 to 7 is implemented.

Images