Fault early warning method and system for variable pitch equipment of wind turbine generator

Abstract

The invention discloses a fault early warning method and system for variable pitch equipment of a wind turbine generator, and relates to the technical field of wind turbine generators, and the fault early warning method comprises the following steps: obtaining operation characteristic parameters of the variable pitch equipment; key factors influencing the operation state of the variable-pitch equipment are recognized; the built variable-pitch equipment fault prediction model is used for predicting the health state of the variable-pitch equipment at the future moment, and the change trend of the health state of the variable-pitch equipment is recognized; and in combination with the operation condition and the safety threshold value of the wind turbine generator pitch control equipment, sending out an early warning signal, and adjusting the operation strategy of the wind turbine generator pitch control equipment. Based on the variable-pitch equipment fault prediction model, the health state of the variable-pitch equipment at the future moment can be accurately predicted, and by recognizing the change trend of the health state, possible problems of the equipment can be recognized in advance, sudden faults are avoided, maintenance measures are taken in advance, and the service life of the equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of wind turbines, and in particular to a fault early warning method and system for variable pitch equipment of a wind turbine. Background Art

[0002] In recent years, the wind power industry has ushered in rapid development, and the installed capacity of wind power has continued to increase. However, compared with this, the operation and maintenance level of wind turbines is still lagging behind. In particular, there is still a lot of room for improvement in technologies such as condition monitoring, fault diagnosis and intelligent maintenance of wind turbines. This is particularly evident compared with the traditional thermal power industry, which has accumulated rich experience and mature technology in these fields. In wind turbines, the variable pitch system, as one of the core components, undertakes a vital task. It can effectively improve the utilization rate of wind energy and ensure stable output power by adjusting the pitch angle of the wind turbine blades in real time, thereby optimizing the operating efficiency of wind turbines.

[0003] However, the existing technology is not convenient for identifying key influencing factors, and cannot help to timely discover potential hidden faults and avoid shutdowns or damage caused by equipment failures, which reduces the reliability and safety of variable pitch equipment. It is also not convenient for accurately predicting the health status of variable pitch equipment at future times, and is not convenient for identifying the changing trend of the health status, so that it is impossible to identify possible problems with the equipment in advance, which increases sudden failures and reduces the service life of the equipment. At the same time, it is not convenient to make dynamic adjustments based on the operating conditions and safety thresholds of the wind turbine set, and it is not convenient to issue early warning signals and adjust the operating strategy.

[0004] Currently, no effective solution has been proposed for the problems in the related technologies. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention proposes a fault warning method and system for wind turbine pitch equipment, which solves the problems mentioned in the above background technology that the prior art is not convenient for identifying key influencing factors, cannot help to timely discover potential fault hazards, avoid shutdown or damage caused by equipment failure, reduce the reliability and safety of the pitch equipment, and is not convenient for accurately predicting the health status of the pitch equipment at future times, and is not convenient for identifying the changing trend of the health status, so as to identify possible problems of the equipment in advance, increase sudden failures, and reduce the service life of the equipment. At the same time, it is not convenient to dynamically adjust in combination with the operating conditions and safety thresholds of the wind turbine, and it is not convenient to issue warning signals and adjust the operating strategy.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions:

[0007] According to one aspect of the present invention, a fault warning method for a wind turbine pitch control device is provided, the fault warning method comprising the following steps:

[0008] S1. Acquire real-time operating parameters of a wind turbine pitch device, and pre-process the real-time operating parameters of the wind turbine pitch device to obtain operating characteristic parameters of the pitch device;

[0009] S2. Analyze the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm to identify the key factors affecting the operating state of the variable pitch device;

[0010] S3. Based on the identified key factors affecting the operation status of the variable pitch equipment, a variable pitch equipment fault prediction model is established, and the health status of the variable pitch equipment at future moments is predicted using the established variable pitch equipment fault prediction model, and the change trend of the health status of the variable pitch equipment is identified;

[0011] S4. Based on the identified changing trend of the health status of the variable pitch equipment, combined with the operating conditions and safety thresholds of the variable pitch equipment of the wind turbine, an early warning signal is issued, and the operating strategy of the variable pitch equipment of the wind turbine is adjusted.

[0012] Further, based on the state analysis algorithm, the obtained operating characteristic parameters of the variable pitch device are analyzed to identify the key factors affecting the operating state of the variable pitch device, including the following steps:

[0013] S21. Evaluate the operating characteristic parameters of the variable pitch equipment using a state analysis algorithm, and based on the evaluation results and characteristic screening rules, screen out a preliminary characteristic parameter subset that affects the change of the operating state as the core parameters;

[0014] S22, initializing parameters of the state analysis algorithm, performing binary conversion on each operating characteristic parameter, and calculating the adaptability score of each operating characteristic parameter based on the correlation between each operating characteristic parameter;

[0015] S23, calculating the step size scaling factor according to the exponential nonlinear decreasing method, and updating the adjustment amplitude and characteristic value of each operating characteristic parameter;

[0016] S24, performing disturbance adjustment on the operation characteristic parameters to increase the diversity of the characteristic parameters;

[0017] S25, calculating the adaptability score of the running characteristic parameters after each adjustment, selecting several characteristic parameters with the best performance as elite characteristic parameters, performing chaos perturbation adjustment on the elite characteristic parameters, recalculating their adaptability scores, and comparing them with the original elite characteristic parameters, and retaining the best elite characteristic parameters;

[0018] S26, judging whether the maximum number of iterations has been reached, if so, outputting the best elite feature as a key factor affecting the operating state of the variable pitch device; if not, returning to step S22 to continue optimization.

[0019] Further, the adaptability score of the running characteristic parameters after each adjustment is calculated, a number of characteristic parameters with the best performance are selected as elite characteristic parameters, the elite characteristic parameters are subjected to chaotic perturbation adjustment, their adaptability scores are recalculated, and compared with the original elite characteristic parameters. The best elite characteristic parameters are retained, including the following steps:

[0020] S261. After each adjustment and optimization, select a number of feature parameters with the best performance to form an elite feature parameter set according to the adaptability score of the running feature parameters, and establish an elite space based on the upper and lower limits of each dimension of the elite feature parameter set;

[0021] S262, based on the chaos mapping method, mapping each characteristic parameter dimension in the elite characteristic parameter set from the elite space to the chaotic space to generate a chaotic characteristic parameter, and performing a chaos mapping operation on each dimension of the chaotic characteristic parameter according to a preset chaotic disturbance probability to generate a new chaotic characteristic parameter;

[0022] S263, based on the inverse chaos mapping method, re-map the new chaotic features back to the elite space to generate a new elite feature parameter set;

[0023] S264. Based on the parameter optimization algorithm, the generated new elite feature parameter set is optimized, and the adaptability score of the optimization result is compared with the adaptability score of the original elite feature parameter, and the elite feature parameter with the highest adaptability score is selected as the best elite feature parameter.

[0024] Further, based on the parameter optimization algorithm, the generated new elite feature parameter set is optimized, and the adaptability score of the optimization result is compared with the adaptability score of the original elite feature parameter, and the elite feature parameter with the highest adaptability score is selected as the best elite feature parameter, including the following steps:

[0025] S2641, determine the size of the elite feature parameter set and set the maximum number of iterations of the parameter optimization algorithm;

[0026] S2642, randomly generating a global exploration point in the parameter space as an initial elite feature parameter to form a set of candidate elite feature parameters in the first column;

[0027] S2643, according to the global exploration rule, combining the randomly generated new elite feature parameters with the parameters in the current elite feature parameter set, generating several new elite feature parameter combinations, and selecting a preset number of candidate elite feature parameters using the golden section method;

[0028] S2644. According to the local optimization rule, the elite feature parameter with the highest adaptability score in the current elite feature parameter set is combined with other elite feature parameters to generate a new set of candidate elite feature parameters;

[0029] S2645, adding the newly generated candidate elite feature parameters to the current elite feature parameter set to form an expanded elite feature parameter set, and sorting the elite feature parameters from high to low according to the adaptability score;

[0030] S2646. Check whether the parameter optimization algorithm has reached the maximum number of iterations. If not, loop through steps S2643 to S2646. If reached, stop the iteration and output the elite feature parameters with the highest adaptability score as the best elite feature parameters.

[0031] Furthermore, the formula of the global exploration rule is:

[0032]

[0033] The formula of the local optimization rule is:

[0034]

[0035] In the formula, Represents the new set of elite feature parameters after global exploration; represents the new elite feature parameter set after local optimization; M a represents the elite feature parameter set in the ath iteration; F a represents the number of elite feature parameters in the ath iteration; X b represents the random elite feature parameter generated by global exploration; X represents the elite feature parameter; X c Represents the elite feature parameter with the highest fitness score selected in the local optimization stage.

[0036] Furthermore, based on the identified key factors affecting the operating state of the variable pitch device, a variable pitch device fault prediction model is established, and the health state of the variable pitch device at a future moment is predicted using the established variable pitch device fault prediction model. Identifying the change trend of the health state of the variable pitch device includes the following steps:

[0037] S31, dividing the identified key factor set affecting the operating state of the pitch control device into a training set and a test set;

[0038] S32, randomly initialize the population of the model optimization algorithm, set the maximum number of iterations and the population size, and set the hyperparameter value range of the variable pitch equipment fault prediction model;

[0039] S33. Use the model optimization algorithm to optimize the hyperparameters of the variable pitch equipment fault prediction model to obtain the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final variable pitch equipment fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

[0040] Furthermore, the hyperparameters of the variable pitch equipment fault prediction model are optimized by using a model optimization algorithm to obtain the optimal hyperparameters, and a final variable pitch equipment fault prediction model is established based on the optimal hyperparameters. The final variable pitch equipment fault prediction model is used to predict the health status of the variable pitch equipment at future moments, and the change trend of the health status of the variable pitch equipment is identified, including the following steps:

[0041] S331, obtaining a hyperparameter subset of the initial variable pitch device fault prediction model, and optimizing the initial hyperparameter subset through an optimization formula to preliminarily screen out hyperparameters related to the variable pitch device fault prediction target;

[0042] S332, selecting several optimal hyperparameter subsets related to the variable pitch device fault prediction target from the hyperparameter subsets of the optimized variable pitch device fault prediction model to form a reference hyperparameter set;

[0043] S333, generating a plurality of reference hyperparameter subsets in the reference hyperparameter set, and synthesizing each reference hyperparameter subset to generate a new hyperparameter subset of the variable pitch equipment fault prediction model;

[0044] S334, merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets based on the subset optimization algorithm as the reference hyperparameter set for the next round of iteration;

[0045] S335. Determine whether the hyperparameter subset of the current variable pitch equipment fault prediction model meets the termination conditions. If not, return to step S334 to continue optimization. If satisfied, output the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

[0046] Further, merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets as the reference hyperparameter set for the next round of iteration based on the subset optimization algorithm includes the following steps:

[0047] S3341, setting a hyperparameter search space of a variable pitch equipment fault prediction model;

[0048] S3342. Randomly initialize a number of hyperparameter sets and feature subsets, where each hyperparameter subset and each feature subset combination represents an optimizer, and the optimizers are combined into an optimizer group;

[0049] S3343. Based on the optimizer group, the variable pitch equipment fault prediction model is trained, and the fitness value of each optimizer is calculated;

[0050] S3344, focusing the optimizers on the currently optimal hyperparameter subset for nearest neighbor search, selecting the optimal hyperparameter subset, calculating its fitness and updating the hyperparameter subset and feature subset;

[0051] S3345. Randomly assign the remaining optimizers to new hyperparameter subsets and feature subsets, perform a global random search, and update the hyperparameters and feature subsets;

[0052] S3346. Repeat steps S3344 and S3345 until the set maximum number of iterations is reached, and use the current optimal hyperparameter subset and feature subset as the reference hyperparameter set for the next round of iteration.

[0053] Furthermore, the optimizer is concentrated on the current optimal hyperparameter subset for neighbor search, and the optimal hyperparameter subset is selected, its fitness is calculated, and the hyperparameter subset and feature subset are updated, including the following steps:

[0054] S33441. Based on the fitness evaluation, the hyperparameter subset with the highest fitness is selected as the current optimal subset;

[0055] S33442, focus the optimizers near the current optimal hyperparameter subset and perform nearest neighbor search;

[0056] S33443. Based on the nearest neighbor search results, a new hyperparameter subset is used to train a variable pitch equipment fault prediction model, and its fitness is calculated;

[0057] S33444. If the fitness of the hyperparameter subset obtained by the nearest neighbor search is greater than the fitness of the current hyperparameter subset, it is used as the new optimal hyperparameter subset, its fitness is calculated, and the hyperparameter subset and feature subset are updated.

[0058] According to another aspect of the present invention, a fault warning system for a wind turbine pitch control device is provided. The fault warning system comprises:

[0059] The data acquisition module is used to obtain the real-time operating parameters of the wind turbine pitch device and pre-process the real-time operating parameters of the wind turbine pitch device to obtain the operating characteristic parameters of the pitch device;

[0060] A state analysis module is used to analyze the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm, and identify key factors affecting the operating state of the variable pitch device;

[0061] The model prediction module is used to establish a fault prediction model for the variable pitch equipment based on the key factors that affect the operating state of the variable pitch equipment, and use the established fault prediction model for the variable pitch equipment to predict the health state of the variable pitch equipment at future moments, and identify the changing trend of the health state of the variable pitch equipment;

[0062] The fault warning module is used to issue a warning signal and adjust the operation strategy of the wind turbine pitch device based on the identified change trend of the health status of the variable pitch device, combined with the operating conditions and safety thresholds of the wind turbine pitch device.

[0063] The beneficial effects of the present invention are:

[0064] 1. The present invention identifies key influencing factors based on the state analysis algorithm, which can help to timely discover potential hidden faults, avoid shutdowns or damage caused by equipment failures, significantly improve the reliability and safety of variable pitch equipment, and establish a variable pitch equipment failure prediction model based on key factors. It can accurately predict the health status of the variable pitch equipment at future moments, and by identifying the changing trend of the health status, it can identify possible problems with the equipment in advance, avoid sudden failures, and take maintenance measures in advance to extend the service life of the equipment. By combining the operating conditions and safety thresholds of the wind turbine for dynamic adjustment, the system can make intelligent decisions based on the actual operating status of the equipment, issue early warning signals and adjust the operating strategy, which not only helps to optimize the operation and management of the wind turbine, but also reduces unnecessary repairs and downtime, thereby reducing maintenance costs.

[0065] 2. The present invention utilizes a state analysis algorithm to gradually screen out key factors that affect the operating state of the variable pitch equipment, ensuring that the retained features are highly relevant and representative, thereby significantly improving the sensitivity and prediction accuracy of the variable pitch equipment fault prediction model to changes in health states. Combined with the exponential nonlinear decreasing method and the chaotic perturbation mechanism, the characteristic values ​​and adaptability scores are dynamically adjusted during parameter screening and optimization, effectively avoiding the local optimal problem that occurs in traditional methods. By accurately screening out key factors, the performance of the variable pitch equipment fault prediction model is significantly enhanced, thereby being able to more timely and accurately identify potential anomalies in the equipment's operating state, providing a reliable basis for subsequent warning signal issuance and adjustment of operating strategies.

[0066] 3. The present invention establishes a variable pitch equipment fault prediction model to accurately predict the health status at future moments, which helps to timely identify the changing trend of the equipment operating status. This capability enables equipment failures to be discovered at an early stage, provides a reliable basis for the issuance of fault warning signals, and significantly shortens the reaction time. In addition, in the process of establishing the variable pitch equipment fault prediction model, the optimization algorithm population is randomly initialized and the model hyperparameters are dynamically optimized, so that the variable pitch equipment fault prediction model can adapt to changes in different operating conditions and external conditions, thereby improving the robustness and adaptability of the variable pitch equipment fault prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0068] Figure 1 is a flow chart of a fault warning method for a wind turbine pitch control device according to an embodiment of the present invention;

[0069] Figure 2 The present invention is a block diagram of a fault warning system for a wind turbine pitch control device according to an embodiment of the present invention.

[0070] In the figure:

[0071] 1. Data acquisition module; 2. State analysis module; 3. Model prediction module; 4. Fault warning module. DETAILED DESCRIPTION

[0072] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0073] In the description of the present invention, unless otherwise specified, the meaning of "plurality" is two or more. In addition, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0074] According to an embodiment of the present invention, a fault warning method and system for a wind turbine pitch control device are provided.

[0075] The present invention is further described with reference to the accompanying drawings and specific embodiments. Figure 1 As shown, according to the fault warning method of the wind turbine pitch control device of the embodiment of the present invention, the fault warning method comprises the following steps:

[0076] S1. Acquire real-time operating parameters of a wind turbine pitch device, and pre-process the real-time operating parameters of the wind turbine pitch device to obtain operating characteristic parameters of the pitch device;

[0077] Specifically, the real-time operating parameters of the wind turbine pitch control equipment include: wind speed, wind direction, generator speed, blade pitch angle, power output, pitch motor current and voltage, hydraulic system pressure and temperature, temperature data, control signals (such as command angle), etc.

[0078] Specifically, the operating characteristic parameters of the variable pitch equipment include: pitch angle change rate, variable pitch system response time, motor load, hydraulic pressure fluctuation, power fluctuation, torque, equipment vibration, hydraulic oil temperature and viscosity, etc.

[0079] S2. Analyze the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm to identify the key factors affecting the operating state of the variable pitch device;

[0080] Specifically, the key factors affecting the operating status of the variable pitch equipment include: environmental factors, mechanical factors, electrical factors, maintenance factors, operational factors, etc.

[0081] S3. Based on the identified key factors affecting the operation status of the variable pitch equipment, a variable pitch equipment fault prediction model is established, and the health status of the variable pitch equipment at future moments is predicted using the established variable pitch equipment fault prediction model, and the change trend of the health status of the variable pitch equipment is identified;

[0082] Specifically, the changing trend of the health status of the variable pitch equipment refers to the dynamic process in which the performance and reliability of the equipment change over time during operation, including: performance degradation trend, failure occurrence trend, maintenance demand trend, key parameter trend, reliability and life trend, etc.

[0083] S4. Based on the identified changing trend of the health status of the variable pitch equipment, combined with the operating conditions and safety thresholds of the variable pitch equipment of the wind turbine, an early warning signal is issued, and the operating strategy of the variable pitch equipment of the wind turbine is adjusted.

[0084] It needs to be explained that real-time operating data is collected, including wind speed, wind direction, temperature, humidity, vibration, pressure, motor current, oil analysis, etc.; sensors and monitoring systems are used to monitor the health status of variable pitch equipment in real time, analyze the collected data, and identify changing trends in the health status; safety thresholds are set according to the equipment manufacturer's guidelines, historical data and industry standards; safety thresholds include but are not limited to maximum vibration levels, temperature limits, pressure limits, etc.; when the monitored data exceeds the preset safety thresholds, the system automatically issues a warning signal; the warning signal can notify the operation and maintenance personnel through sound, light, text message or email; conduct in-depth analysis of the potential causes behind the warning signal to determine the specific type and location of the fault; according to the warning signal and fault diagnosis results, adjust the operating parameters of the variable pitch equipment, such as reducing the amplitude of the blade angle change, reducing the speed, etc.

[0085] Preferably, based on the state analysis algorithm, analyzing the obtained operating characteristic parameters of the variable pitch device, identifying the key factors affecting the operating state of the variable pitch device includes the following steps:

[0086] S21. Evaluate the operating characteristic parameters of the variable pitch equipment using a state analysis algorithm, and based on the evaluation results and characteristic screening rules, screen out a preliminary characteristic parameter subset that affects the change of the operating state as the core parameters;

[0087] S22, initializing parameters of the state analysis algorithm, performing binary conversion on each operating characteristic parameter, and calculating the adaptability score of each operating characteristic parameter based on the correlation between each operating characteristic parameter;

[0088] S23, calculating the step size scaling factor according to the exponential nonlinear decreasing method, and updating the adjustment amplitude and characteristic value of each operating characteristic parameter;

[0089] S24, performing disturbance adjustment on the operation characteristic parameters to increase the diversity of the characteristic parameters;

[0090] S25, calculating the adaptability score of the running characteristic parameters after each adjustment, selecting several characteristic parameters with the best performance as elite characteristic parameters, performing chaos perturbation adjustment on the elite characteristic parameters, recalculating their adaptability scores, and comparing them with the original elite characteristic parameters, and retaining the best elite characteristic parameters;

[0091] S26, judging whether the maximum number of iterations has been reached, if so, outputting the best elite feature as a key factor affecting the operating state of the variable pitch device; if not, returning to step S22 to continue optimization.

[0092] In order to facilitate understanding of the above technical solutions of the present invention, the present invention will now analyze the operating characteristic parameters of the variable pitch device obtained based on the state analysis algorithm in the actual process, and identify the key factors affecting the operating state of the variable pitch device in detail.

[0093] For example, when analyzing the variable pitch device of a wind turbine, the goal is to find out the key factors that affect the operating state of the device through the state analysis algorithm. The operating characteristic parameters of the variable pitch device include wind speed, current, blade angle, device temperature, etc. Then input the collected data, as shown in Table 1:

[0094] Table 1: Collected data

[0095]

[0096] Step 1: Feature evaluation and preliminary screening

[0097] 1) Calculate the correlation between each feature and the target variable (such as the label of the device failure state or device operation state):

[0098] For example, the target variable is the equipment operating status (1 for normal, 0 for fault), and there is corresponding data. First, calculate the correlation between each feature (wind speed, current, blade angle, equipment temperature) and the target variable. For example, through the Pearson correlation coefficient or other correlation measurement methods, the following results are obtained:

[0099] Correlation between wind speed and equipment operating status: 0.85.

[0100] Correlation between current and equipment operating status: 0.65.

[0101] Correlation between blade angle and equipment operating status: 0.90.

[0102] Correlation between device temperature and device operating status: 0.80.

[0103] 2) Calculate the redundancy between features:

[0104] Use Pearson correlation coefficient or mutual information to calculate the redundancy between each pair of features. For example, the redundancy is calculated as follows:

[0105] Redundancy of wind speed and current: 0.75.

[0106] Redundancy of wind speed and blade angle: 0.60.

[0107] Redundancy between wind speed and equipment temperature: 0.65.

[0108] Redundancy of current and blade angle: 0.70.

[0109] Redundancy of current and device temperature: 0.50.

[0110] Redundancy of blade angle and equipment temperature: 0.60.

[0111] 3) Apply the minimum redundancy maximum correlation algorithm to filter features:

[0112] The minimum redundancy maximum relevance method is used to select the most relevant features with the least redundancy by weighing the relevance and redundancy. The minimum redundancy maximum relevance algorithm selects features by optimizing the objective function:

[0113]

[0114] Where f(D) represents the objective function, which is used to calculate the score of the feature parameter set D; S represents a subset of the feature parameter set (such as wind speed, current, blade angle, device temperature, etc.); |S| represents the number of features in the subset S of the feature parameter set; x i represents the i-th feature in the subset S of the feature parameter set; y represents the target variable (usually the operating status or fault status of the equipment, assuming that 1 represents normal and 0 represents fault, etc.); cor(x i , y) represents x i The correlation between y and y (e.g., Pearson correlation coefficient).

[0115] Based on the optimization objective function, the following optimal feature set can be obtained:

[0116] Blade angle (correlation 0.90) and wind speed (correlation 0.85) are selected as preliminary core features because they have high correlation and low redundancy.

[0117] Step 2: Initialize the algorithm parameters

[0118] 1) Initialize the parameters of the improved krill swarm algorithm:

[0119] Particle group size: 50.

[0120] Maximum number of iterations: 100.

[0121] Initial step size: 0.1.

[0122] Convergence factor: 0.9.

[0123] 2) Binary conversion:

[0124] The initially selected characteristic parameters (blade angle, wind speed) are binary-encoded for processing in the krill swarm algorithm. For example:

[0125] Wind speed (10.5 m / s) is represented as 1010 in binary.

[0126] The blade angle (34°) is represented in binary as 100010.

[0127] 3) Initialize the position and velocity of the particle:

[0128] Assume that each particle represents a feature selection scheme, and each feature (wind speed, blade angle) has two possible states (selected or not selected). Initialize the positions of the particle swarm (each particle corresponds to a binary vector) and set its velocity.

[0129] Step 3: Calculation of Adaptability Score

[0130] 1) Fitness function:

[0131] In each iteration, the fitness of each particle is calculated, and the fitness function is usually based on the relevance and redundancy of the features, as well as the relevance of the target variable (device state) to the selected features. For example, suppose the fitness function is as follows:

[0132] f(D)=α·cor(S,y)·∑ i，j∈S cor(x i , x j );

[0133] Where α and β both represent weight factors; f(D) represents the objective function, which is used to calculate the score of the feature parameter set D; S represents a subset of the feature parameter set (such as wind speed, current, blade angle, equipment temperature, etc.); x i represents the i-th feature in the subset S of the feature parameter set; y represents the target variable (usually the operating status or fault status of the equipment, assuming that 1 represents normal and 0 represents fault, etc.); cor(x i , y) represents x i The correlation between y and y (e.g., Pearson correlation coefficient).

[0134] 2) Update particle position and velocity: Based on the fitness function and velocity update rule, update the velocity and position of each particle.

[0135] Step 4: Disturbance Adjustment and Diversity Increase

[0136] 1) Perturbation adjustment: Perturb the particle positions to increase the diversity of the search space. For example, the feature selection scheme can be changed by randomly changing the bits in the binary code.

[0137] 2) Select the particles with the best fitness: After each iteration, the particles with the best fitness scores are selected as elite particles.

[0138] Step 5: Elite feature selection and chaos perturbation

[0139] 1) Chaotic perturbation: Perform chaotic perturbations on elite particles to avoid falling into local optimal solutions. For example, chaotic mapping is used to update the positions of elite particles.

[0140] 2) Comparison of elite particles with original particles: Recalculate the adaptability score of elite particles after chaos adjustment and compare it with the original particles, and select particles with better adaptability as the final solution.

[0141] Step 6: Iteration and termination conditions

[0142] 1) Check whether the maximum number of iterations has been reached: If the number of iterations has reached the preset maximum value (100 times), the current optimal feature is output as the key factor. If not, return to step S22 to continue optimization.

[0143] 2) Output the optimal features: After 100 iterations, assuming that blade angle and wind speed are selected as the final optimal features, these two features are output as the key factors affecting the operating status of the variable pitch equipment.

[0144] Specifically, the state analysis algorithm is a two-stage hybrid feature selection algorithm based on minimum redundancy maximum correlation and improved krill swarm algorithm, that is, the minimum redundancy maximum correlation algorithm is used to evaluate the features. In order to screen out a highly correlated and low-redundant gene subset, and then, the improved krill swarm algorithm is combined to iteratively optimize the features. The krill swarm algorithm is a meta-heuristic swarm intelligence algorithm that simulates the process of krill swarms responding to specific dangers and environments and gradually increasing population density. In order to enable the krill swarm algorithm to handle feature selection problems, the present invention uses the coding conversion principle to discretize it, and in order to further improve the global search capability of the krill swarm algorithm, it is improved by nonlinear step size reduction and elite particle local search.

[0145] Preferably, calculating the adaptability score of the running characteristic parameters after each adjustment, selecting several characteristic parameters with the best performance as elite characteristic parameters, performing chaotic perturbation adjustment on the elite characteristic parameters, recalculating their adaptability scores, and comparing them with the original elite characteristic parameters, retaining the best elite characteristic parameters includes the following steps:

[0146] S261. After each adjustment and optimization, select a number of feature parameters with the best performance to form an elite feature parameter set according to the adaptability score of the running feature parameters, and establish an elite space based on the upper and lower limits of each dimension of the elite feature parameter set;

[0147] S262, based on the chaos mapping method, mapping each characteristic parameter dimension in the elite characteristic parameter set from the elite space to the chaotic space to generate a chaotic characteristic parameter, and performing a chaos mapping operation on each dimension of the chaotic characteristic parameter according to a preset chaotic disturbance probability to generate a new chaotic characteristic parameter;

[0148] S263, based on the inverse chaos mapping method, re-map the new chaotic features back to the elite space to generate a new elite feature parameter set;

[0149] S264. Based on the parameter optimization algorithm, the generated new elite feature parameter set is optimized, and the adaptability score of the optimization result is compared with the adaptability score of the original elite feature parameter, and the elite feature parameter with the highest adaptability score is selected as the best elite feature parameter.

[0150] It should be explained that the contribution of each feature is first evaluated, usually measured by the performance of the model (such as accuracy, F1 score, etc.). The fitness scores of these features reflect their performance in the current model; according to the fitness scores, several features with the best performance are selected to form an elite feature set. For example, assuming there are multiple features, the features with the top scores are selected as elite features; after selecting the elite features, we set reasonable upper and lower limits for each feature, which define the elite space. The elite space is the effective range of the feature parameters, which is used to limit the adjustment range of the parameters during the optimization process; the chaos mapping method is used to map each feature parameter in the elite feature set from the elite space to a wider chaotic space. Chaotic mapping generates new feature values ​​in a nonlinear and random way, making the adjusted features more diverse and avoiding over-reliance on the current optimal features; a perturbation probability is set to control the degree to which each feature is perturbed. This means that some features may undergo larger adjustments, while other features may remain unchanged. This process increases the search space and helps avoid the algorithm from falling into the local optimum; through the perturbation operation, the numerical range of the features is disrupted and redistributed to generate a new chaotic feature set; the adjusted chaotic features are remapped back to the elite space. Inverse chaos mapping is to restore the feature values ​​in the chaotic space back to the valid range of the elite space through a specific algorithm. In this way, the features remain in the elite space, but their values ​​have been adjusted by perturbation; after inverse chaos mapping, a new elite feature set is obtained, which has been adjusted but still meets the restrictions of the elite space; the new feature set after inverse chaos mapping is optimized using parameter optimization algorithms (such as particle swarm optimization, genetic algorithm, Fibonacci tree optimization algorithm, etc.). The goal of this process is to maximize the fitness score of the model by searching for new feature combinations; the optimized new feature set will produce a new fitness score. This score is compared with the fitness score of the original elite feature set to determine whether better results have been achieved; if the optimized feature set performs better (higher fitness score), it is retained as the new elite feature set. If the optimization result fails to improve the score, the original elite feature set is retained.

[0151] Specifically, the parameter optimization algorithm is the Fibonacci tree optimization algorithm (FTO), which is a computational intelligent optimization algorithm based on the Fibonacci optimization principle and the golden section method. It constructs a Fibonacci tree structure through alternating iterations of global exploration and local optimization to solve the optimization problem.

[0152] Preferably, based on the parameter optimization algorithm, the generated new elite feature parameter set is optimized, and the adaptability score of the optimization result is compared with the adaptability score of the original elite feature parameter, and the elite feature parameter with the highest adaptability score is selected as the best elite feature parameter, which includes the following steps:

[0153] S2641, determine the size of the elite feature parameter set and set the maximum number of iterations of the parameter optimization algorithm;

[0154] S2642, randomly generating a global exploration point in the parameter space as an initial elite feature parameter to form a set of candidate elite feature parameters in the first column;

[0155] S2643, according to the global exploration rule, combining the randomly generated new elite feature parameters with the parameters in the current elite feature parameter set, generating several new elite feature parameter combinations, and selecting a preset number of candidate elite feature parameters using the golden section method;

[0156] S2644. According to the local optimization rule, the elite feature parameter with the highest adaptability score in the current elite feature parameter set is combined with other elite feature parameters to generate a new set of candidate elite feature parameters;

[0157] S2645, adding the newly generated candidate elite feature parameters to the current elite feature parameter set to form an expanded elite feature parameter set, and sorting the elite feature parameters from high to low according to the adaptability score;

[0158] S2646. Check whether the parameter optimization algorithm has reached the maximum number of iterations. If not, loop through steps S2643 to S2646. If reached, stop the iteration and output the elite feature parameters with the highest adaptability score as the best elite feature parameters.

[0159] Preferably, the formula of the global exploration rule is:

[0160]

[0161] The formula of the local optimization rule is:

[0162]

[0163] In the formula, Represents the new set of elite feature parameters after global exploration; represents the new elite feature parameter set after local optimization; M a represents the elite feature parameter set in the ath iteration; F a represents the number of elite feature parameters in the ath iteration; X b represents the random elite feature parameter generated by global exploration; X represents the elite feature parameter; X c Represents the elite feature parameter with the highest fitness score selected in the local optimization stage.

[0164] Preferably, based on the identified key factors affecting the operating state of the pitch device, a pitch device fault prediction model (i.e., a long short-term memory neural network LSTM) is established, and the established pitch device fault prediction model is used to predict the health state of the pitch device at future moments, and identifying the change trend of the health state of the pitch device includes the following steps:

[0165] S31, dividing the identified key factor set affecting the operating state of the pitch control device into a training set and a test set;

[0166] S32, randomly initialize the population of the model optimization algorithm, set the maximum number of iterations and the population size, and set the hyperparameter value range of the variable pitch equipment fault prediction model;

[0167] S33. Use the model optimization algorithm to optimize the hyperparameters of the variable pitch equipment fault prediction model to obtain the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final variable pitch equipment fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

[0168] It needs to be explained that by analyzing the operation process of the variable pitch equipment, the key factors affecting the health status of the equipment are identified, such as motor temperature, vibration frequency, current and voltage fluctuations, blade angle, wind speed, etc.; historical data related to the above key factors are collected and sorted, and the data sources may include real-time data collected by sensors, maintenance records, operation logs, etc.; the collected influencing factor data sets are divided into training sets and test sets according to a certain proportion. Usually, the training set is used to train the model, and the test set is used to evaluate the performance of the model. For example, 80% of the data is used for training and 20% of the data is used for testing; in order to optimize the hyperparameters of the variable pitch equipment fault prediction model, a model optimization algorithm is selected, such as particle swarm optimization, genetic algorithm, ant colony algorithm, decentralized search algorithm, etc. These algorithms can efficiently find the optimal hyperparameter combination in a large search space. In the optimization process, we first randomly initialize the algorithm population, and each individual represents a potential hyperparameter combination. In this way, the optimization algorithm will explore different hyperparameter values ​​to find the parameter configuration that best suits the model; determine the maximum number of iterations and population size of the algorithm. The maximum number of iterations limits the running time of the optimization process, while the population size determines the number of hyperparameter combinations involved in the optimization in each generation. For example, the maximum number of iterations is set to 100 and the population size is 50; during the optimization process, we need to set the value range for the hyperparameters of the variable pitch equipment fault prediction model. These hyperparameters may include the learning rate of the model, the regularization parameter, the depth of the tree, etc. By setting a reasonable value range, the optimization algorithm can explore the best parameters within these ranges; the hyperparameters are optimized using the selected model optimization algorithm. In each round of optimization, the algorithm trains the model based on the hyperparameter combination in the current population and evaluates its prediction performance (such as accuracy, F1 score, AUC, etc.). The algorithm continuously adjusts the hyperparameters to find the hyperparameter combination with the best performance; after completing all iterations, the model optimization algorithm will obtain a set of optimal hyperparameter combinations. These hyperparameters enable the variable pitch equipment fault prediction model to show the best prediction performance on the training set; the optimal hyperparameters obtained by optimization are used to establish the final variable pitch equipment fault prediction model. The model can predict the probability of equipment failure based on input features (such as motor temperature, vibration frequency, etc.), and can evaluate the health status of the equipment in real time; use the final prediction model to predict the health status of the variable pitch equipment at future moments. The model will predict whether the equipment is likely to fail and the health trend of the equipment in the future based on the input real-time monitoring data; based on the prediction results, analyze the changing trend of the equipment health status. If the prediction results show that the health status has declined or the risk of failure has increased, relevant personnel can take preventive measures in advance, such as adjusting equipment operation, troubleshooting, or arranging maintenance work.

[0169] Specifically, the model optimization algorithm is a decentralized search algorithm, which is an evolutionary algorithm based on heuristics. The basic idea is: to build a set of reference solutions according to the principle of diversity, and then obtain a better new solution set through a series of selection, combination, and improvement operations. The most obvious difference between the decentralized search algorithm and the genetic algorithm is that its series of operation strategies are no longer based on the principle of randomness, but use a decentralized-convergence-aggregation intelligent iterative mechanism.

[0170] Long Short-Term Memory Neural Network (LSTM) Long Short-Term Memory Network (LSTM) is a special recurrent neural network (RNN) that introduces memory units based on traditional RNN, enabling it to better capture long-term dependencies when processing time series data.

[0171] Preferably, the hyperparameters of the variable pitch device fault prediction model are optimized by using a model optimization algorithm to obtain the optimal hyperparameters, and a final variable pitch device fault prediction model is established based on the optimal hyperparameters. The final variable pitch device fault prediction model is used to predict the health state of the variable pitch device at a future moment, and identifying the change trend of the health state of the variable pitch device includes the following steps:

[0172] S331, obtaining a hyperparameter subset of the initial variable pitch device fault prediction model, and optimizing the initial hyperparameter subset through an optimization formula to preliminarily screen out hyperparameters related to the variable pitch device fault prediction target;

[0173] S332, selecting several optimal hyperparameter subsets related to the variable pitch device fault prediction target from the hyperparameter subsets of the optimized variable pitch device fault prediction model to form a reference hyperparameter set;

[0174] S333, generating a plurality of reference hyperparameter subsets in the reference hyperparameter set, and synthesizing each reference hyperparameter subset to generate a new hyperparameter subset of the variable pitch equipment fault prediction model;

[0175] S334, merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets based on the subset optimization algorithm as the reference hyperparameter set for the next round of iteration;

[0176] S335. Determine whether the hyperparameter subset of the current variable pitch equipment fault prediction model meets the termination conditions. If not, return to step S334 to continue optimization. If satisfied, output the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

[0177] Preferably, merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets as the reference hyperparameter set for the next round of iteration based on the subset optimization algorithm includes the following steps:

[0178] S3341, setting a hyperparameter search space of a variable pitch equipment fault prediction model;

[0179] S3342. Randomly initialize a number of hyperparameter sets and feature subsets, where each hyperparameter subset and each feature subset combination represents an optimizer, and the optimizers are combined into an optimizer group;

[0180] S3343. Based on the optimizer group, the variable pitch equipment fault prediction model is trained, and the fitness value of each optimizer is calculated;

[0181] S3344, focusing the optimizers on the currently optimal hyperparameter subset for nearest neighbor search, selecting the optimal hyperparameter subset, calculating its fitness and updating the hyperparameter subset and feature subset;

[0182] S3345. Randomly assign the remaining optimizers to new hyperparameter subsets and feature subsets, perform a global random search, and update the hyperparameters and feature subsets;

[0183] S3346. Repeat steps S3344 and S3345 until the set maximum number of iterations is reached, and use the current optimal hyperparameter subset and feature subset as the reference hyperparameter set for the next round of iteration.

[0184] Specifically, the subset optimization algorithm is the bee algorithm, which is a new cluster intelligent search algorithm that finds the optimal solution to the problem by simulating the foraging behavior of bees in nature. An optimization algorithm proposed. In the algorithm, the location of the food source represents a set of solutions to the optimization problem. The location of the food source and its immediate neighborhood form a small search area, called a flower cluster. In the process of searching for food, the bee colony first sends out scout bees to search. The scout bees continuously fly from one food source to another to search for all possible food sources. When the profitability of the searched food source (such as the sugar content) exceeds a certain amount, the scout bee returns to the hive, unloads the honey and dances a magical "swing dance" on the "dance board". This "swing dance" is a necessary tool for the bee colony to communicate information, which enables the entire bee colony to understand the location of the discovered food source, the distance to the hive and the profitability. After dancing, the scout bee leads the follower bees waiting outside the hive to fly back to the corresponding food source to collect honey, and the food source with high profitability will recruit more follower bees to collect honey. During the honey-gathering process, the bees calculate the profitability of the food source they are collecting honey from, and return to the hive to perform a "swing dance" to convey the information of the current food source profitability to the bee colony. If the food source profitability is still high, more follower bees will be recruited to collect honey; if the food source profitability is low, the current food source will be abandoned.

[0185] Preferably, the optimizer is concentrated on the currently optimal hyperparameter subset for nearest neighbor search, the optimal hyperparameter subset is selected, its fitness is calculated, and the hyperparameter subset and feature subset are updated, including the following steps:

[0186] S33441. Based on the fitness evaluation, the hyperparameter subset with the highest fitness is selected as the current optimal subset;

[0187] S33442, focus the optimizers near the current optimal hyperparameter subset and perform nearest neighbor search;

[0188] S33443. Based on the nearest neighbor search results, a new hyperparameter subset is used to train a variable pitch equipment fault prediction model, and its fitness is calculated;

[0189] S33444. If the fitness of the hyperparameter subset obtained by the nearest neighbor search is greater than the fitness of the current hyperparameter subset, it is used as the new optimal hyperparameter subset, its fitness is calculated, and the hyperparameter subset and feature subset are updated.

[0190] According to another embodiment of the present invention, Figure 2 As shown, a fault warning system for a wind turbine pitch control device is also provided, and the fault warning system includes:

[0191] The data acquisition module 1 is used to acquire the real-time operating parameters of the wind turbine pitch device and pre-process the real-time operating parameters of the wind turbine pitch device to obtain the operating characteristic parameters of the pitch device;

[0192] The state analysis module 2 is used to analyze the obtained operation characteristic parameters of the variable pitch device based on the state analysis algorithm, and identify the key factors affecting the operation state of the variable pitch device;

[0193] Model prediction module 3 is used to establish a variable pitch device fault prediction model based on the identified key factors affecting the operating state of the variable pitch device, and use the established variable pitch device fault prediction model to predict the health state of the variable pitch device at future moments, and identify the change trend of the health state of the variable pitch device;

[0194] The fault warning module 4 is used to issue a warning signal and adjust the operation strategy of the wind turbine pitch device based on the identified change trend of the health status of the pitch device, combined with the operating conditions and safety thresholds of the wind turbine pitch device.

[0195] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A fault warning method for a wind turbine pitch control device, characterized in that: The fault warning method comprises the following steps: S1. Acquire real-time operating parameters of a wind turbine pitch device, and pre-process the real-time operating parameters of the wind turbine pitch device to obtain operating characteristic parameters of the pitch device; S2. Analyze the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm to identify the key factors affecting the operating state of the variable pitch device; S3. Based on the identified key factors affecting the operation status of the variable pitch equipment, a variable pitch equipment fault prediction model is established, and the health status of the variable pitch equipment at future moments is predicted using the established variable pitch equipment fault prediction model, and the change trend of the health status of the variable pitch equipment is identified; S4. Based on the identified changing trend of the health status of the variable pitch equipment, combined with the operating conditions and safety thresholds of the variable pitch equipment of the wind turbine, an early warning signal is issued, and the operating strategy of the variable pitch equipment of the wind turbine is adjusted.

2. A fault warning method for wind turbine pitch control equipment according to claim 1, characterized in that: The method of analyzing the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm and identifying the key factors affecting the operating state of the variable pitch device includes the following steps: S21. Evaluate the operating characteristic parameters of the variable pitch equipment using a state analysis algorithm, and based on the evaluation results and characteristic screening rules, screen out a preliminary characteristic parameter subset that affects the change of the operating state as the core parameters; S22, initializing parameters of the state analysis algorithm, performing binary conversion on each operating characteristic parameter, and calculating the adaptability score of each operating characteristic parameter based on the correlation between each operating characteristic parameter; S23, calculating the step size scaling factor according to the exponential nonlinear decreasing method, and updating the adjustment amplitude and characteristic value of each operating characteristic parameter; S24, performing disturbance adjustment on the operation characteristic parameters to increase the diversity of the characteristic parameters; S25, calculating the adaptability score of the running characteristic parameters after each adjustment, selecting several characteristic parameters with the best performance as elite characteristic parameters, performing chaos perturbation adjustment on the elite characteristic parameters, recalculating their adaptability scores, and comparing them with the original elite characteristic parameters, and retaining the best elite characteristic parameters; S26, judging whether the maximum number of iterations has been reached, if so, outputting the best elite feature as a key factor affecting the operating state of the variable pitch device; if not, returning to step S22 to continue optimization.

3. A fault warning method for wind turbine pitch control equipment according to claim 2, characterized in that: The step of calculating the adaptability score of the running characteristic parameters after each adjustment, selecting several characteristic parameters with the best performance as elite characteristic parameters, performing chaotic perturbation adjustment on the elite characteristic parameters, recalculating their adaptability scores, and comparing them with the original elite characteristic parameters, and retaining the best elite characteristic parameters includes the following steps: S261. After each adjustment and optimization, select a number of feature parameters with the best performance to form an elite feature parameter set according to the adaptability score of the running feature parameters, and establish an elite space based on the upper and lower limits of each dimension of the elite feature parameter set; S262, based on the chaos mapping method, mapping each characteristic parameter dimension in the elite characteristic parameter set from the elite space to the chaotic space to generate a chaotic characteristic parameter, and performing a chaos mapping operation on each dimension of the chaotic characteristic parameter according to a preset chaotic disturbance probability to generate a new chaotic characteristic parameter; S263, based on the inverse chaos mapping method, re-map the new chaotic features back to the elite space to generate a new elite feature parameter set; S264. Based on the parameter optimization algorithm, the generated new elite feature parameter set is optimized, and the adaptability score of the optimization result is compared with the adaptability score of the original elite feature parameter, and the elite feature parameter with the highest adaptability score is selected as the best elite feature parameter.

4. A fault warning method for wind turbine pitch control equipment according to claim 3, characterized in that: The method of optimizing the newly generated elite feature parameter set based on the parameter optimization algorithm, comparing the adaptability score of the optimization result with the adaptability score of the original elite feature parameter, and selecting the elite feature parameter with the highest adaptability score as the best elite feature parameter includes the following steps: S2641, determine the size of the elite feature parameter set and set the maximum number of iterations of the parameter optimization algorithm; S2642, randomly generating a global exploration point in the parameter space as an initial elite feature parameter to form a set of candidate elite feature parameters in the first column; S2643, according to the global exploration rule, combining the randomly generated new elite feature parameters with the parameters in the current elite feature parameter set, generating several new elite feature parameter combinations, and selecting a preset number of candidate elite feature parameters using the golden section method; S2644. According to the local optimization rule, the elite feature parameter with the highest adaptability score in the current elite feature parameter set is combined with other elite feature parameters to generate a new set of candidate elite feature parameters; S2645, adding the newly generated candidate elite feature parameters to the current elite feature parameter set to form an expanded elite feature parameter set, and sorting the elite feature parameters from high to low according to the adaptability score; S2646. Check whether the parameter optimization algorithm has reached the maximum number of iterations. If not, loop through steps S2643 to S2646. If reached, stop the iteration and output the elite feature parameters with the highest adaptability score as the best elite feature parameters.

5. A fault warning method for wind turbine pitch control equipment according to claim 4, characterized in that: The formula of the global exploration rule is: The formula of the local optimization rule is: In the formula, Represents the new set of elite feature parameters after global exploration; represents the new elite feature parameter set after local optimization; M a Represents the elite feature parameter set in the a-th iteration; F a represents the number of elite feature parameters in the ath iteration; X b represents the random elite feature parameter generated by global exploration; X represents the elite feature parameter; X c Represents the elite feature parameter with the highest fitness score selected in the local optimization stage.

6. A fault warning method for wind turbine pitch control equipment according to claim 1, characterized in that: The method of establishing a fault prediction model for a pitch device based on the identified key factors affecting the operating state of the pitch device, and using the established fault prediction model for the pitch device to predict the health state of the pitch device at a future moment, and identifying the change trend of the health state of the pitch device includes the following steps: S31, dividing the identified key factor set affecting the operating state of the pitch control device into a training set and a test set; S32, randomly initialize the population of the model optimization algorithm, set the maximum number of iterations and the population size, and set the hyperparameter value range of the variable pitch equipment fault prediction model; S33. Use the model optimization algorithm to optimize the hyperparameters of the variable pitch equipment fault prediction model to obtain the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final variable pitch equipment fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

7. A fault warning method for wind turbine pitch control equipment according to claim 6, characterized in that: The method of optimizing the hyperparameters of the variable pitch device fault prediction model by using the model optimization algorithm to obtain the optimal hyperparameters, and establishing the final variable pitch device fault prediction model based on the optimal hyperparameters, and using the final variable pitch device fault prediction model to predict the health state of the variable pitch device at a future moment, and identifying the change trend of the health state of the variable pitch device includes the following steps: S331, obtaining a hyperparameter subset of the initial variable pitch device fault prediction model, and optimizing the initial hyperparameter subset through an optimization formula to preliminarily screen out hyperparameters related to the variable pitch device fault prediction target; S332, selecting several optimal hyperparameter subsets related to the variable pitch device fault prediction target from the hyperparameter subsets of the optimized variable pitch device fault prediction model to form a reference hyperparameter set; S333, generating a plurality of reference hyperparameter subsets in the reference hyperparameter set, and synthesizing each reference hyperparameter subset to generate a new hyperparameter subset of the variable pitch equipment fault prediction model; S334, merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets based on the subset optimization algorithm as the reference hyperparameter set for the next round of iteration; S335. Determine whether the hyperparameter subset of the current variable pitch equipment fault prediction model meets the termination conditions. If not, return to step S334 to continue optimization. If satisfied, output the optimal hyperparameters, and establish the final variable pitch equipment fault prediction model based on the optimal hyperparameters. Use the final fault prediction model to predict the health status of the variable pitch equipment at future times, and identify the changing trend of the health status of the variable pitch equipment.

8. A fault warning method for wind turbine pitch control equipment according to claim 7, characterized in that: The step of merging the current reference hyperparameter set with the newly generated hyperparameter subset of the variable pitch equipment fault prediction model, and selecting several optimal hyperparameter subsets as the reference hyperparameter sets for the next iteration based on a subset optimization algorithm comprises the following steps: S3341, setting a hyperparameter search space of a variable pitch equipment fault prediction model; S3342. Randomly initialize a number of hyperparameter sets and feature subsets, where each hyperparameter subset and each feature subset combination represents an optimizer, and the optimizers are combined into an optimizer group; S3343. Based on the optimizer group, the variable pitch equipment fault prediction model is trained, and the fitness value of each optimizer is calculated; S3344, focusing the optimizers on the currently optimal hyperparameter subset for nearest neighbor search, selecting the optimal hyperparameter subset, calculating its fitness and updating the hyperparameter subset and feature subset; S3345. Randomly assign the remaining optimizers to new hyperparameter subsets and feature subsets, perform a global random search, and update the hyperparameters and feature subsets; S3346. Repeat steps S3344 and S3345 until the set maximum number of iterations is reached, and use the current optimal hyperparameter subset and feature subset as the reference hyperparameter set for the next round of iteration.

9. A fault warning method for wind turbine pitch control equipment according to claim 8, characterized in that: The process of focusing the optimizers on the currently optimal hyperparameter subset for neighbor search, selecting the optimal hyperparameter subset, calculating its fitness, and updating the hyperparameter subset and feature subset includes the following steps: S33441. Based on the fitness evaluation, the hyperparameter subset with the highest fitness is selected as the current optimal subset; S33442, focus the optimizers near the current optimal hyperparameter subset and perform nearest neighbor search; S33443. Based on the nearest neighbor search results, a new hyperparameter subset is used to train a variable pitch equipment fault prediction model, and its fitness is calculated; S33444. If the fitness of the hyperparameter subset obtained by the nearest neighbor search is greater than the fitness of the current hyperparameter subset, it is used as the new optimal hyperparameter subset, its fitness is calculated, and the hyperparameter subset and feature subset are updated.

10. A fault warning system for a wind turbine pitch control device, used to implement the fault warning method for a wind turbine pitch control device according to any one of claims 1 to 9, characterized in that: The fault warning system includes: The data acquisition module is used to obtain the real-time operating parameters of the wind turbine pitch device and pre-process the real-time operating parameters of the wind turbine pitch device to obtain the operating characteristic parameters of the pitch device; A state analysis module is used to analyze the obtained operating characteristic parameters of the variable pitch device based on the state analysis algorithm, and identify key factors affecting the operating state of the variable pitch device; The model prediction module is used to establish a fault prediction model for the variable pitch equipment based on the key factors that affect the operating state of the variable pitch equipment, and use the established fault prediction model for the variable pitch equipment to predict the health state of the variable pitch equipment at future moments, and identify the changing trend of the health state of the variable pitch equipment; The fault warning module is used to issue a warning signal and adjust the operation strategy of the wind turbine pitch device based on the identified change trend of the health status of the variable pitch device, combined with the operating conditions and safety thresholds of the wind turbine pitch device.

Images