Intelligent speed regulation method and system for a thermal power plant fan

Abstract

The application provides a kind of intelligent speed regulation method and system of thermal power station fan, method includes: based on the extraction of time-frequency characteristics of multi-modal data, based on the extraction of dynamic working condition characteristics of boiler working condition parameters, fusion is obtained joint state vector;Based on the generation of fan continuous action instruction of joint state vector;With non-dominated sorting genetic algorithm based on the optimization of continuous action instruction of boiler working condition parameters, generate optimized control solution set and execute speed regulation.The application realizes real-time sensing of operating state by multi-modal data fusion and dynamic working condition adaptive optimization mechanism, first extracts features to generate joint state vector, then provides model support by combining online updating aerodynamic knowledge base;Finally, through working condition adaptive optimization algorithm, dynamically adjust the weight coefficient of wind energy efficiency and mechanical load, impose exponential penalty on thrust coefficient constraint, generate optimized control solution set that matches boiler load demand, coal quality characteristics and meets safety boundary, solve the control mismatch problem of traditional method.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of fan speed regulation, in particular to a method and system for intelligent speed regulation of a fan in a thermal power plant. BACKGROUND

[0002] Current fan speed regulation in thermal power plants mainly adopts PID control, model predictive control (MPC) or traditional genetic algorithm optimization strategy. PID control relies on fixed parameters and is difficult to adapt to dynamic conditions such as wind speed fluctuations and coal quality changes; MPC can handle multiple objective constraints, but requires an accurate aerodynamic model to support, and factors such as blade wear and dust accumulation during actual fan operation can cause model mismatch. Traditional genetic algorithms (such as NSGA-II) can generate a set of multi-objective solutions, but the initial population is usually randomly generated without incorporating the condition adaptability data from the prior knowledge base, and the fitness function mostly uses static weights, which cannot dynamically adjust the optimization direction according to real-time parameters such as boiler load and coal calorific value.

[0003] The existing methods have three defects: first, the multi-modal operating data (such as vibration, speed, and wind speed time series data) and the boiler operating parameters (load, coal calorific value) are not effectively integrated, resulting in a disconnection between the control command and the actual operating state; second, the aerodynamic model update relies on offline calibration and cannot dynamically correct key parameters such as circulation decay and turbulence intensity through online learning mechanisms, resulting in a continuous decline in model accuracy over a long period of operation; third, the fitness function of the optimization algorithm does not embed the safety boundary constraints of the thrust coefficient, and only uses a simple elimination strategy for individuals that violate the constraints, which can easily lead to the optimization solution set deviating from the engineering safety requirements. SUMMARY

[0004] The present application aims to at least solve the technical problem of the lack of adaptability of control strategies to dynamic conditions in the prior art, and particularly innovatively proposes a method and system for intelligent speed regulation of a fan in a thermal power plant.

[0005] In order to achieve the above-mentioned purpose of the present application, the present application provides a method for intelligent speed regulation of a fan in a thermal power plant, the method comprising:

[0006] S1, collecting multi-modal data of the fan operation and boiler operating parameters, the multi-modal data including wind speed data, vibration data, speed data, load data and time series data of coal quality;

[0007] S2, extracting multi-modal time-frequency features based on the multi-modal data of the fan operation, extracting dynamic operating condition features based on the boiler operating parameters, and performing feature fusion on the dynamic operating condition features and the multi-modal time-frequency features to obtain a joint state vector;

[0008] S3, establishing a prior knowledge base of fan aerodynamic characteristics and a dynamic control strategy of speed and pitch angle, and generating a continuous action command for the fan based on the joint state vector;

[0009] S4, optimizing the continuous action instruction based on the boiler working condition parameter and using a non-dominated sorting genetic algorithm to generate an optimized control solution set;

[0010] S5, issuing the optimized control solution set to a fan controller and using the fan controller to execute the optimized control solution set to regulate the speed of the fan.

[0011] In another aspect, the present application also provides a power plant fan intelligent speed regulation system, and the power plant fan intelligent speed regulation method comprises the following steps:

[0012] A data acquisition module is configured to acquire fan operation multi-modal data and boiler working condition parameters.

[0013] A preprocessing module is connected to the data acquisition module and configured to preprocess the fan operation multi-modal data and boiler working condition parameters and extract effective feature information.

[0014] A feature extraction module is connected to the preprocessing module and configured to extract a feature vector reflecting fan aerodynamic characteristics and operating states based on the preprocessed data.

[0015] A control strategy module is connected to the feature extraction module and configured to generate fan speed and pitch angle adjustment instructions according to the extracted feature vector and the power plant fan intelligent speed regulation method.

[0016] An execution module is connected to the control strategy module and configured to receive and execute the speed and pitch angle adjustment instructions to adjust the fan operating state.

[0017] A feedback optimization module is connected to the execution module and the data acquisition module and configured to acquire executed fan operation data, construct a feedback optimization closed loop, dynamically adjust fan control strategy parameters, and continuously optimize fan operation performance.

[0018] The beneficial effects of the present application: the present application can realize real-time sensing of the running state by multi-modal data fusion and dynamic working condition adaptive optimization mechanism, first extract the time-frequency features of the fan vibration, speed, wind speed and other time series data based on step S2, and fuse with the working condition parameters such as boiler load and coal calorific value to generate a joint state vector, realize real-time sensing of the running state; then combined with the online updating aerodynamic knowledge base (including full working condition aerodynamic parameter table and control rule set) constructed in step S3, provide accurate model support for the control strategy; finally, through the working condition adaptive NSGA-II optimization algorithm (non-dominated sorting genetic algorithm) in step S4, dynamically adjust the weight coefficients of wind energy efficiency and mechanical load, and impose exponential penalty on the thrust coefficient constraint, generate an optimization control solution set that matches the current boiler load demand, coal calorific value characteristics, and meets the safety boundary, so as to solve the control mismatch problem caused by the dynamic change of working condition in the traditional method, realize the improvement of wind energy capture efficiency and the reduction of mechanical load fluctuation.

[0019] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0020] The above and / or additional aspects and advantages of the present application will become apparent and be readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which:

[0021] Figure 1 is a flow chart of a method for intelligent speed regulation of a power plant fan. DETAILED DESCRIPTION

[0022] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, in which the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as limiting the present application.

[0023] Example 1

[0024] As shown in Figure 1 , a method for intelligent speed regulation of a power plant fan, the method comprising:

[0025] S1, collecting fan running multi-modal data and boiler working condition parameters, the multi-modal data including wind speed data, vibration data, speed data, load data and time series data of coal quality;

[0026] In step S1, it should be noted that collecting multimodal data on fan operation and boiler operating parameters is the first step in implementing the intelligent speed control method. The accuracy and completeness of this data directly affect the precision and efficiency of subsequent steps. Wind speed data reflects real-time changes in the external environment and is crucial for adjusting the fan speed and blade pitch angle. Vibration data reflects the mechanical condition of the fan blades, preventing potential failures. Speed ​​data is directly related to the fan's energy conversion efficiency, while load data and coal quality time-series data provide information on the actual operating status of the boiler.

[0027] During the data acquisition process, high-precision sensors and data acquisition cards are used to ensure the real-time nature and accuracy of the data. Simultaneously, the acquired data undergoes preprocessing to remove noise and outliers, improving data quality. Preprocessing steps include data cleaning, filtering, and normalization.

[0028] S2. Extract multimodal time-frequency features based on the multimodal data of the fan operation, extract dynamic operating condition features based on the boiler operating condition parameters, and fuse the dynamic operating condition features with the multimodal time-frequency features to obtain a joint state vector.

[0029] In step S2, by fusing the multimodal time-frequency characteristics of the fan operation with the dynamic characteristics of the boiler operating conditions, a comprehensive reflection of the actual operating status of the fan and changes in the external environment can be achieved. During the feature extraction stage, signal processing algorithms, such as wavelet transform and empirical mode decomposition, are used to extract representative multimodal time-frequency features from the raw data. These features can capture the fan's vibration at different frequencies, speed fluctuations, and wind speed trends, thus accurately reflecting the fan's aerodynamic performance and mechanical state. Simultaneously, dynamic operating condition features extracted based on boiler operating parameters, such as load fluctuations and changes in coal calorific value, reflect the boiler's actual operating requirements and changes in external conditions.

[0030] In the feature fusion stage, multimodal time-frequency features are organically combined with dynamic operating condition features to form a joint state vector containing rich information. This step fully utilizes the complementarity between different features, improving the accuracy and comprehensiveness of state perception.

[0031] S3. Establish a prior knowledge base of wind turbine aerodynamic characteristics and a dynamic control strategy for speed and pitch angle, and generate continuous action commands for the wind turbine based on the joint state vector.

[0032] In step S3, it is crucial to emphasize that establishing a prior knowledge base of the wind turbine's aerodynamic characteristics is key to ensuring the effectiveness of the control strategy. This knowledge base contains aerodynamic parameter tables and control rule sets for all operating conditions, providing accurate model support for the dynamic control strategy. The aerodynamic parameter tables record detailed wind turbine performance data, such as power coefficient and thrust coefficient, at different wind speeds and pitch angles. The control rule set, based on extensive engineering experience and experimental results, defines the control logic and adjustment strategies under different operating conditions.

[0033] When generating continuous action commands, the system first determines the optimal speed and pitch angle adjustment targets under the current operating conditions based on the joint state vector and the aerodynamic characteristic prior knowledge base. Then, combining the dynamic control strategy of speed and pitch angle, it calculates the continuous adjustment commands that satisfy the action space constraints.

[0034] S4. Based on the boiler operating parameters and using a non-dominated sorting genetic algorithm, optimize the continuous action commands to generate an optimized control solution set;

[0035] In step S4, the Non-Dominated Sorting Genetic Algorithm (NSGA-II) is used to optimize continuous action commands, aiming to find a control strategy that achieves the optimal balance between wind energy efficiency and mechanical load. This algorithm iteratively improves the solution population by simulating the process of natural selection, with each generation evolving towards better wind energy capture efficiency and lower mechanical load fluctuations.

[0036] During the optimization process, the algorithm first defines a multi-objective optimization problem based on the boiler operating parameters, which includes two core objectives: maximizing wind energy efficiency and minimizing mechanical load. To ensure the safety and feasibility of the solution, a thrust coefficient safety boundary constraint is also incorporated to prevent the wind turbine from operating beyond the safe range.

[0037] The algorithm distinguishes individuals in the population through non-dominated sorting, ranking them according to their performance in the multi-objective space. In each generation, the algorithm selects the top-ranked individuals as parents and generates offspring through simulated binary crossover and polynomial mutation operations. This process iterates until a preset maximum number of iterations or a convergence condition is reached, ultimately obtaining an optimized set of control solutions.

[0038] This set of control solutions represents the optimal adjustment strategies for fan speed and pitch angle under different boiler operating conditions. By selecting the sequence of action commands that best matches the current load and coal calorific value, the system can achieve precise control over the fan's operating status, thereby improving wind energy capture efficiency, reducing mechanical load fluctuations, extending fan life, and improving the overall operating efficiency of the thermal power plant.

[0039] S5. The optimized control solution set is sent to the fan controller, and the fan controller is used to execute the optimized control solution set to adjust the speed of the fan.

[0040] In step S5, after receiving the optimized control solution set, the fan controller immediately parses and executes these instructions. Each instruction in the control solution set specifies the adjustment amount of the fan speed and pitch angle to ensure that the fan can respond quickly and accurately to the current boiler operating conditions. During execution, the fan controller monitors the fan's operating status in real time, including key parameters such as speed, pitch angle, vibration, and energy consumption, to ensure that the adjustment instructions are effectively executed and that the fan operates within a safe and stable range.

[0041] Meanwhile, to further improve the accuracy and response speed of wind turbine speed regulation, this invention also employs a closed-loop feedback optimization mechanism. After executing the speed regulation command, the system collects real-time operating data of the wind turbine through the data acquisition module, including key indicators such as air volume, energy consumption, and vibration. This data is then input into the Bayesian optimization framework, combining historical and real-time data to dynamically adjust the wind turbine control strategy parameters. The Bayesian optimization framework uses a Gaussian process model as a prior distribution, intelligently selecting the next evaluation point through the acquisition function, iteratively updating the model parameters until the convergence condition or the preset number of iterations is reached. This process continuously optimizes the wind turbine's control strategy, making it more adaptable to the current operating environment and working conditions.

[0042] In summary, the intelligent speed control method for thermal power plant fans in this embodiment first achieves comprehensive perception of the fan operating status and the external environment by collecting multimodal data on fan operation and boiler operating parameters. Based on the data acquisition, the system further extracts multimodal time-frequency features and dynamic operating condition features, and fuses these features to form a joint state vector. This step improves the accuracy and comprehensiveness of the state perception.

[0043] Subsequently, the system established a priori knowledge base of the wind turbine's aerodynamic characteristics and a dynamic control strategy for speed and pitch angle. This knowledge base and control strategy provided precise model support and logical basis for generating continuous action commands. When generating commands, the system fully considered the optimal speed and pitch angle adjustment targets under the current operating conditions, as well as the constraints of the action space, ensuring the feasibility and effectiveness of the commands.

[0044] In optimizing continuous action commands, this invention employs a non-dominated sorting genetic algorithm (NSGA-II). This algorithm iteratively improves the solution population by simulating the process of natural selection, ultimately obtaining a set of optimized control solutions. This set of solutions represents the optimal adjustment strategies for fan speed and blade pitch angle under different boiler operating conditions.

[0045] Finally, the system sends the optimized control solution set to the fan controller for execution. During execution, the fan controller monitors the fan's operating status in real time and continuously adjusts the control strategy parameters dynamically through a closed-loop feedback optimization mechanism. This process ensures that the fan can respond quickly and accurately to the current boiler operating conditions, while improving wind energy capture efficiency and reducing mechanical load fluctuations.

[0046] As an optional embodiment of the present invention, optionally, the expression for obtaining the joint state vector in step S2 is:

[0047]

[0048]

[0049]

[0050]

[0051]

[0052] in, Indicates the time-frequency characteristics of the vibration signal. The scale parameter represents the wavelet transform. This represents the translation parameter of the wavelet transform. This represents the raw vibration signal acquired by the triaxial accelerometer. Represents a time variable. Indicates conjugate. This represents the statistical feature vector of wind speed data. This represents the mean. This represents wind speed data collected by lidar and ultrasonic anemometers. Indicates standard deviation, This indicates taking the maximum value. This indicates taking the minimum value. This represents the statistical feature vector of rotational speed data. This represents the fan speed data measured by the encoder. This represents a multimodal feature fusion vector. This represents the mean vector of the training dataset. This represents the standard deviation vector of the training dataset. This represents the joint state vector.

[0053] As an optional embodiment of the present invention, the establishment of the prior knowledge base of the aerodynamic characteristics of the fan in step S3 may include:

[0054] S301. Construct a hybrid aerodynamic model of the wind turbine based on its geometric parameters and operating conditions, and use the hybrid aerodynamic model to perform fusion simulation to obtain fusion simulation data;

[0055] The method for constructing the hybrid aerodynamic model of the wind turbine in step S301 involves combining a computational fluid dynamics (CFD) model with a data-driven machine learning model. The CFD model can simulate the aerodynamic performance of the wind turbine under different conditions, such as power coefficient and thrust coefficient, based on the wind turbine's geometric parameters and operating range. The data-driven machine learning model, on the other hand, can learn the wind turbine's operating patterns and characteristics from a large amount of historical data, improving prediction accuracy. By fusing these two models, more comprehensive and accurate aerodynamic characteristic data of the wind turbine can be obtained. In the fusion simulation phase, the CFD model is used to perform simulation calculations on different geometric parameters and operating conditions, obtaining a series of aerodynamic performance data. Simultaneously, these simulation data are compared and verified with the prediction results of the machine learning model, continuously optimizing the model's parameters and structure to improve the model's prediction accuracy and generalization ability. Finally, the verified fusion simulation data is stored in the aerodynamic characteristic prior knowledge base.

[0056] S302. Based on the hybrid aerodynamic model, design a wind tunnel experiment, collect aerodynamic parameters under different working conditions, and obtain experimental datasets through static and dynamic calibration.

[0057] The purpose of designing the wind tunnel experiment in step S302 is to verify the accuracy and reliability of the hybrid aerodynamic model. In the experiment, different operating conditions such as wind speed, blade pitch angle, and rotational speed are simulated to collect aerodynamic parameters of the wind turbine under different states, such as power, thrust, and torque. To ensure the accuracy and comparability of the experimental data, static and dynamic calibrations are required. Static calibration mainly calibrates the accuracy and zero-point drift of the measuring equipment to ensure data accuracy; while dynamic calibration calibrates the measuring equipment under different operating conditions to eliminate the influence of changes in operating conditions on the measurement results. Through this series of experimental and calibration steps, a rich and accurate experimental dataset is obtained. After obtaining the experimental dataset, it is compared and analyzed with the fused simulation data to further verify the accuracy and reliability of the hybrid aerodynamic model. Simultaneously, based on the feedback from the experimental data, necessary adjustments and optimizations are made to the model to improve its prediction accuracy and generalization ability.

[0058] S303. Based on the fused simulation data and experimental dataset, generate an aerodynamic parameter table and control rule set covering all operating conditions, and define the pitch angle adjustment range and thrust coefficient safety boundary in combination with empirical rules.

[0059] In step S303, the aerodynamic parameter table records detailed aerodynamic performance data of the wind turbine under different combinations of wind speed, pitch angle, and rotational speed, such as power coefficient and thrust coefficient. Based on these aerodynamic parameter tables, the control rule set defines the control logic and adjustment strategies for different operating conditions to ensure stable operation of the wind turbine under various conditions while maximizing wind energy capture efficiency. Furthermore, combining engineering experience and experimental results, the adjustment range of the pitch angle is defined to ensure that the wind turbine does not exceed the safe range during adjustment. Simultaneously, to further improve safety, a safety boundary for the thrust coefficient is defined to prevent the wind turbine from overloading under extreme operating conditions.

[0060] S304. Verify the accuracy of the hybrid aerodynamic model based on the aerodynamic parameter table and control rule set. Compare the predicted values ​​with the measured values ​​through simulation and field testing. Dynamically update the hybrid aerodynamic model using an online learning mechanism.

[0061] In step S304, the accuracy verification of the hybrid aerodynamic model is a crucial step in ensuring its effectiveness in practical applications. This step primarily assesses the model's accuracy by comparing simulation predictions with field measured values. During verification, the hybrid aerodynamic model is first simulated using aerodynamic parameter tables and control rule sets to obtain a series of predicted values. These predicted values ​​are then compared and analyzed with the measured values ​​obtained from field testing to evaluate the model's prediction accuracy and error range. If there is a significant deviation between the predicted and measured values, it indicates that the model's predictive ability under certain operating conditions needs improvement. In this case, an online learning mechanism can be used to dynamically update the hybrid aerodynamic model. The online learning mechanism can iteratively optimize the model using new measured data, continuously improving the model's prediction accuracy and generalization ability. Through this series of verification and update steps, it is ensured that the hybrid aerodynamic model maintains high accuracy and reliability in practical applications.

[0062] S305. Based on the hybrid aerodynamic model, output the updated prior knowledge base of the wind turbine aerodynamic characteristics, which includes the verified aerodynamic parameter table, control rule set and anomaly diagnosis library, and generate a knowledge base version log to record update information.

[0063] The updated aerodynamic characteristic prior knowledge base for the wind turbine, as described in step S305, is a core component of the intelligent speed control system. It not only includes rigorously validated aerodynamic parameter tables and control rule sets, but also adds an anomaly diagnosis library for quickly identifying and resolving potential faults or anomalies during wind turbine operation. The aerodynamic parameter tables and control rule sets provide the system with precise control guidelines under different operating conditions, ensuring the efficient and stable operation of the wind turbine. The anomaly diagnosis library, by collecting and analyzing historical fault data, establishes a correspondence between fault characteristics and solutions, enabling the system to respond quickly and take effective corrective measures when a fault occurs, reducing the impact of the fault on the operation of the thermal power plant.

[0064] The knowledge base version log records detailed information for each update, including the updated content, update time, and personnel involved. This logging mechanism helps system administrators track the latest status of the knowledge base, ensuring that all relevant personnel can access the latest knowledge and information in a timely manner.

[0065] By continuously updating and improving the prior knowledge base of aerodynamic characteristics of wind turbines, the intelligent speed regulation system for thermal power plant wind turbines of this invention can continuously adapt to new operating conditions and environments, improve the control accuracy and response speed of wind turbines, further extend the service life of wind turbines, and improve the overall operating efficiency of thermal power plants.

[0066] As an optional embodiment of the present invention, optionally, the expression of the dynamically updated hybrid aerodynamic model in S304 is:

[0067]

[0068]

[0069]

[0070] in, This indicates the updated blade circulation. This represents the initial circulation value. This represents the circulation decay rate as it changes over time. This represents the time step of discretization. Indicates the learning rate. Indicates the total number of operating points. Indicates the first CFD flow field velocity vector at each working point Represents the local time integral variable. Indicates the first BEM flow field velocity vectors at each operating point This indicates the updated turbulence intensity. Indicates the initial turbulence intensity. This represents the turbulence intensity adjustment factor. Indicates the number of experimental operating points. Indicates the first The experimental lift coefficient at each working point Indicates the first Simulated lift coefficients at various operating points Indicates the initial decay rate. Indicates the acceleration factor of the decay rate. Indicates the cumulative running time. This indicates the design life of the wind turbine blades.

[0071] As an optional embodiment of the present invention, the dynamic control strategy for speed and pitch angle established in step S3 may include:

[0072] S306. Define control objectives based on the joint state vector and the prior knowledge base of wind turbine aerodynamic characteristics. The control objectives include maximizing wind energy capture efficiency, minimizing mechanical load fluctuations, and satisfying the thrust coefficient safety boundary.

[0073] Maximizing wind energy capture efficiency in step S306 means that the wind turbine needs to adjust its rotational speed and pitch angle according to the current wind speed and operating conditions to capture as much wind energy as possible, thereby improving power generation efficiency. Simultaneously, minimizing mechanical load fluctuations helps reduce wear and fatigue of the wind turbine, extending its service life. Furthermore, meeting the thrust coefficient safety boundary is crucial to ensuring the wind turbine operates within a safe range, preventing structural damage or safety accidents caused by excessive thrust.

[0074] S307. Construct a strategy network based on the joint state vector, and use the strategy network to obtain continuous speed and pitch angle adjustment commands that satisfy the action space constraints.

[0075] In step S307, the strategy network is a deep learning-based model that extracts key information from the joint state vector and generates corresponding speed and pitch angle adjustment commands based on the control objective. To ensure the feasibility and effectiveness of these commands in practical applications, the strategy network design considers constraints in the action space, such as the adjustment range and speed of speed and pitch angle. By training and optimizing the strategy network, it can quickly and accurately generate appropriate adjustment commands under different operating conditions and environments, thereby achieving intelligent speed control of the wind turbine. After obtaining the adjustment commands, they are sent to the wind turbine's control system for execution, realizing dynamic adjustment of the wind turbine's speed and pitch angle.

[0076] The method for constructing the policy network is as follows: First, a large amount of historical operating data is collected, including key parameters such as wind speed, engine speed, pitch angle, and corresponding power output under different operating conditions. This data will be used as training samples to train the policy network. Next, the collected data is preprocessed, including data cleaning and normalization, to improve data quality and training efficiency. Then, the architecture of the policy network is designed, including input layers, hidden layers, and output layers. The input layer receives the joint state vector as input, the hidden layer performs nonlinear transformations on the input data through a multi-layer neural network to extract key features, and the output layer generates continuous speed and pitch angle adjustment commands that satisfy the action space constraints. After determining the network architecture, a suitable loss function and optimization algorithm are selected for network training. The loss function is used to measure the gap between the network output and the actual target, and the optimization algorithm is used to continuously adjust the network parameters to minimize the loss function. Through multiple iterations of training, until the network converges or reaches the preset number of training rounds, the trained policy network is finally obtained. This policy network can quickly and accurately generate appropriate speed and pitch angle adjustment commands according to different operating conditions and requirements, realizing intelligent speed control of the wind turbine.

[0077] S308. Combining the aerodynamic parameter table and control rule set in the prior knowledge base, and based on the control objective, dynamically correcting the original continuous rotational speed and pitch angle adjustment commands output by the strategy network. The correction includes circulation attenuation compensation and turbulence intensity adjustment.

[0078] The purpose of dynamic correction in step S308 is to further improve the accuracy and applicability of the speed and pitch angle adjustment commands output by the strategy network. Due to various uncertainties and interference factors in the actual operating environment, such as wind speed fluctuations and changes in turbulence intensity, these factors may cause a certain deviation between the output of the strategy network and actual requirements. Therefore, it is necessary to dynamically correct the original continuous speed and pitch angle adjustment commands by combining the aerodynamic parameter table and control rule set in the prior knowledge base, in order to compensate for these deviations, ensure that the wind turbine can operate stably under various operating conditions, and maximize wind energy capture efficiency.

[0079] Circulation decay compensation primarily considers the circulation decay caused by factors such as airflow friction and blade deformation during wind turbine blade operation. This decay affects the aerodynamic performance of the wind turbine, thereby influencing the adjustment requirements for its speed and pitch angle. Therefore, it is necessary to compensate the speed and pitch angle adjustment commands output by the strategy network based on circulation decay data in the prior knowledge base to eliminate the impact of circulation decay on wind turbine performance.

[0080] Turbulence intensity adjustment considers the impact of turbulence intensity on wind turbine performance. Changes in turbulence intensity lead to variations in the wind turbine's aerodynamic load and power output, thus affecting the adjustment requirements for speed and pitch angle. Therefore, it is necessary to dynamically adjust the speed and pitch angle adjustment commands output by the strategy network, based on turbulence intensity data from a priori knowledge base, to adapt to changes in turbulence intensity and ensure stable operation of the wind turbine under various turbulence intensities. By combining circulation attenuation compensation with turbulence intensity adjustment, the accuracy and applicability of the speed and pitch angle adjustment commands output by the strategy network can be further improved, achieving intelligent speed control of the wind turbine.

[0081] S309. Generate continuous action instructions that the wind turbine controller can execute based on the corrected continuous speed and pitch angle adjustment instructions, wherein the continuous action instructions include the time sequence of speed adjustment amount and pitch angle adjustment amount.

[0082] The corrected continuous speed and pitch angle adjustment commands in step S309 are further processed and converted to generate continuous action commands that the wind turbine controller can directly execute. These commands are given in the form of a time sequence, detailing the specific values ​​of the speed and pitch angle that the wind turbine needs to adjust at different time points. This design ensures that the wind turbine control system can respond accurately and promptly to adjustment needs, achieving precise control of the wind turbine speed and pitch angle. After generating the continuous action commands, they are sent to the wind turbine control system for execution. The control system adjusts the wind turbine's operating state according to the commands to meet current operating conditions and goals. Through this series of control steps, the intelligent speed regulation system for thermal power plant wind turbines of this invention can achieve intelligent speed regulation control of wind turbine speed and pitch angle, improve the control accuracy and response speed of the wind turbine, further extend the service life of the wind turbine, and improve the overall operating efficiency of the thermal power plant.

[0083] As an optional embodiment of the present invention, optionally, the expression for dynamically correcting the original continuous rotational speed and pitch angle adjustment command output by the strategy network in step S308 is:

[0084] circulation attenuation compensation term:

[0085]

[0086]

[0087] in, This indicates the speed adjustment amount after circulation attenuation compensation. This represents the original speed adjustment amount output by the strategy network. This represents the circulation compensation coefficient. Indicates wind speed With pitch angle The corresponding circulation decay rate, This represents the baseline value of the compensation coefficient below the rated wind speed. This indicates the compensation attenuation rate after the wind speed exceeds the rated value. Indicates the rated wind speed of the fan;

[0088] Turbulence intensity adjustment item:

[0089]

[0090]

[0091] in, This indicates the amount of pitch angle adjustment after adjusting for turbulence intensity. This represents the original pitch angle adjustment amount output by the policy network. Indicates the turbulence adjustment factor. Indicates wind speed With pitch angle The corresponding experimental turbulence intensity, Indicates wind speed With pitch angle The corresponding simulated turbulence intensity, This represents the pitch angle sensitivity coefficient. This represents the optimal pitch angle.

[0092] As an optional embodiment of the present invention, optionally, generating the optimized control solution set in step S4 includes:

[0093] S401. Based on the boiler operating parameters and the continuous operation commands of the fan, a multi-objective optimization problem is constructed. The optimization problem takes maximizing wind energy efficiency and minimizing mechanical load as its core objectives, and incorporates thrust coefficient safety boundary constraints.

[0094] The purpose of constructing the multi-objective optimization problem in step S401 is to find an optimal control strategy that maximizes the wind energy capture efficiency of the wind turbine while minimizing the mechanical load, under the premise of satisfying the thrust coefficient safety boundary constraint. To achieve this goal, boiler operating parameters and continuous operation commands of the wind turbine are used as inputs to construct an optimization function containing multiple objectives. The wind energy efficiency objective function measures the wind turbine's wind energy capture capability under different control strategies, while the mechanical load objective function evaluates the mechanical load level of the wind turbine under different control strategies. By solving this multi-objective optimization problem, a set of optimized control solutions that satisfy the constraints can be obtained. These solutions represent the optimal control strategies that the wind turbine should adopt under different operating conditions.

[0095] S402. Based on the multi-objective optimization problem, the continuous action instructions are incorporated when initializing the population using a non-dominated sorting genetic algorithm.

[0096] In step S402, a non-dominated sorting genetic algorithm is used to solve the multi-objective optimization problem. This algorithm is an optimization algorithm based on natural selection and genetic mechanisms, capable of finding a set of non-dominated solutions, i.e., the Pareto optimal solution set, in complex multi-objective problems. During population initialization, the continuous operation commands of the wind turbine are incorporated into the population as part of the initial solution, which helps the algorithm converge to the optimal solution more quickly. Through continuous selection, crossover, and mutation operations, the algorithm can gradually optimize the solutions in the population, making them closer to the Pareto front. During the solution process, the algorithm also evaluates the solutions according to the wind energy efficiency objective function and the mechanical load objective function, and sorts the solutions according to the non-dominated relationship, thereby ensuring that the final optimized control solution set satisfies both the thrust coefficient safety boundary constraint and maximizes wind energy capture efficiency and minimizes mechanical load.

[0097] S403. Define a fitness function based on the boiler operating parameters and the aerodynamic parameter table, dynamically adjust the target weight of wind energy capture efficiency using the fitness function, and impose an exponential penalty on individuals that violate the thrust coefficient constraint to generate a fitness evaluation system that meets the operating requirements.

[0098] In step S403, the fitness function is a key indicator used to evaluate the quality of each solution in the population under specific operating conditions. In this invention, the fitness function is defined based on boiler operating parameters and aerodynamic parameter tables, and can comprehensively consider multiple aspects such as wind energy capture efficiency, mechanical load, and thrust coefficient safety boundary. By dynamically adjusting the target weight of wind energy capture efficiency, the fitness function can balance the relationship between wind energy capture and mechanical load under different operating conditions, ensuring that the wind turbine maintains structural safety while operating efficiently.

[0099] Simultaneously, for individuals that violate the thrust coefficient constraint, the fitness function applies an exponential penalty to increase their fitness value and reduce their competitiveness in the population. This design helps to gradually eliminate solutions that do not meet safety constraints during the solution process, guiding the algorithm to converge towards the optimal solution set that meets the operating conditions. By constructing a fitness evaluation system that meets the operating conditions, this invention can achieve a comprehensive evaluation and optimization of wind turbine control strategies, further improving the operating efficiency and safety of thermal power plant wind turbines.

[0100] S404. Perform iterative optimization of genetic operations based on the initialized population and fitness function. Select high-quality individuals through non-dominated sorting, and generate offspring by combining simulated binary crossover and polynomial mutation until the maximum number of iterations is reached, and obtain the iterative optimization result.

[0101] In step S404, based on the initialized population and fitness function, genetic operations are initiated for iterative optimization. During this process, non-dominated sorting is used to select high-performing individuals in each generation, those exhibiting excellent performance in multiple aspects such as wind energy capture efficiency, mechanical load, and thrust coefficient safety boundary. To generate new offspring, this invention combines simulated binary crossover and polynomial mutation. Simulated binary crossover mimics natural hybridization, generating new offspring by exchanging partial genes of parent individuals; this process helps maintain population diversity and explore new solution spaces. Polynomial mutation, on the other hand, makes small, random adjustments to the genes of offspring individuals to increase population variability and avoid premature convergence. By continuously iterating these genetic operations, the solutions in the population gradually converge to the Pareto front, i.e., a set of non-dominated solutions that satisfies the thrust coefficient safety boundary constraint while maximizing wind energy capture efficiency and minimizing mechanical load. When the preset maximum number of iterations is reached, the iterative optimization process ends, and the obtained iterative optimization result is the desired optimized control solution set.

[0102] S405. Based on the boiler operating parameters and iterative optimization results, output the control solution set after non-dominated sorting, and select the action command sequence that best matches the current load and coal calorific value to generate the optimized control solution set.

[0103] In step S405, based on boiler operating parameters and the control solution set obtained through iterative optimization and non-dominated sorting, the system further filters out the action command sequence that best matches the current operating conditions of the thermal power plant, such as load and coal calorific value. This filtering process comprehensively considers various factors, including the fan operating status, boiler requirements, and external environment, ensuring that the selected control solution set can achieve optimal operating results under the current conditions. By generating the optimized control solution set, this invention provides more precise and efficient guidance for the intelligent speed regulation control of thermal power plant fans, further improving the overall operating efficiency and safety of the thermal power plant. In practical applications, this optimized control solution set will be directly applied to the fan control system to achieve intelligent adjustment of fan speed and pitch angle, thereby meeting the ever-changing operating needs of the thermal power plant.

[0104] As an optional embodiment of the present invention, optionally, the expression of the fitness function in step S402 is:

[0105]

[0106] in, Represents the fitness function. This represents the action instruction vector corresponding to an individual. This indicates the dynamic weighting of wind energy efficiency. Indicates wind energy capture efficiency. Represents the dynamic weight of mechanical load. Indicates mechanical load fluctuation. This represents the coefficient of the penalty term. This indicates the thrust coefficient under the current action command.

[0107] As an optional embodiment of the present invention, the method may further include:

[0108] S6. Establish a feedback optimization closed loop, input the air volume, energy consumption and vibration data after execution into the Bayesian optimization framework, and dynamically adjust the fan control strategy parameters by combining historical data and real-time data; the Bayesian optimization framework uses a Gaussian process model as a prior distribution, selects the next evaluation point through the acquisition function, and iteratively updates the model parameters until the convergence condition or the preset number of iterations is reached.

[0109] The purpose of establishing a feedback optimization closed loop in step S6 is to continuously optimize the wind turbine control strategy and improve the operating efficiency and stability of the thermal power plant. During actual operation, data such as wind volume, energy consumption, and vibration of the wind turbine are collected in real time and input into the Bayesian optimization framework. This framework uses a Gaussian process model as a prior distribution and can dynamically adjust the wind turbine control strategy parameters based on historical and real-time data. By selecting the next evaluation point through the acquisition function and iteratively updating the model parameters, the Bayesian optimization framework can gradually approach the optimal combination of control strategy parameters. When the convergence condition is met or the preset number of iterations is reached, the optimization process ends, and the obtained control strategy parameters are the optimal solution. This feedback optimization closed loop design enables the intelligent speed regulation system for thermal power plant wind turbines of this invention to continuously learn and adapt to different operating conditions, further improving the control accuracy and response speed of the wind turbine.

[0110] Example 2

[0111] A smart speed control system for thermal power plant fans, and a smart speed control method for thermal power plant fans: the system further includes:

[0112] The data acquisition module is used to collect multi-modal data of fan operation and boiler operating parameters;

[0113] The preprocessing module, connected to the acquisition module, is used to preprocess the multimodal data of the fan operation and the boiler operating parameters, and extract effective feature information.

[0114] The feature extraction module, connected to the preprocessing module, is used to extract feature vectors reflecting the aerodynamic characteristics and operating status of the fan based on the preprocessed data.

[0115] The control strategy module, connected to the feature extraction module, is used to generate speed and pitch angle adjustment commands for the wind turbine based on the extracted feature vector and the intelligent speed regulation method for the thermal power plant wind turbine.

[0116] An execution module, connected to the control strategy module, is used to receive and execute the speed and pitch angle adjustment commands to adjust the wind turbine operating status.

[0117] The feedback optimization module, connected to the execution module and the data acquisition module, is used to collect the wind turbine operation data after execution, construct a feedback optimization closed loop, and dynamically adjust the wind turbine control strategy parameters to continuously optimize the wind turbine operation performance.

[0118] It should be noted that the intelligent speed control system for thermal power plant fans in this embodiment is used to implement the intelligent speed control method for thermal power plant fans in Embodiment 1. This system achieves intelligent adjustment of the fan speed and pitch angle through the collaborative work of various modules. The data acquisition module is responsible for real-time acquisition of fan operating data and boiler operating parameters. The preprocessing module performs preprocessing operations such as cleaning and filtering on the acquired raw data to extract effective feature information and reduce the impact of noise on subsequent analysis. The feature extraction module further extracts feature vectors reflecting the aerodynamic characteristics and operating state of the fan based on the preprocessed data. These feature vectors are the key basis for formulating control strategies. The control strategy module generates fan speed and pitch angle adjustment commands based on the extracted feature vectors and the intelligent speed control method for thermal power plant fans described in Embodiment 1. These commands are sent to the fan through the execution module to achieve precise adjustment of the fan operating state. Finally, the feedback optimization module collects the executed fan operating data, constructs a feedback optimization closed loop, and dynamically adjusts the fan control strategy parameters using advanced technologies such as the Bayesian optimization framework to continuously optimize the fan's operating performance. Through this closed-loop optimization process, the system can continuously learn and adapt to different operating conditions, further improving the control accuracy, response speed, and overall operating efficiency and stability of the wind turbine.

[0119] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent speed regulation of a fan in a thermal power plant, characterized in that, The method comprises: S1, collecting fan operation multi-modal data and boiler operating condition parameters, the multi-modal data including wind speed data, vibration data, rotational speed data, load data and time series data of coal quality; S2, extracting multi-modal time-frequency features based on the fan operation multi-modal data, extracting dynamic operating condition features based on the boiler operating condition parameters, and performing feature fusion on the dynamic operating condition features and the multi-modal time-frequency features to obtain a joint state vector; S3, establishing a fan aerodynamic characteristic prior knowledge base and a rotational speed and pitch angle dynamic control strategy, and generating a continuous action instruction of the fan based on the joint state vector; In step S3, the fan aerodynamic characteristic prior knowledge base is established, which comprises: S301, constructing a hybrid aerodynamic model of the fan based on fan geometric parameters and operating condition ranges, performing fusion simulation using the hybrid aerodynamic model, and obtaining fusion simulation data; S302, designing a wind tunnel experiment based on the hybrid aerodynamic model, collecting aerodynamic parameters under different operating conditions, and obtaining an experimental data set through static and dynamic calibration; S303, generating an aerodynamic parameter table and a control rule set covering all operating conditions based on the fusion simulation data and the experimental data set, and defining a pitch angle adjustment range and a thrust coefficient safety boundary in combination with experience rules; S304, verifying the accuracy of the hybrid aerodynamic model based on the aerodynamic parameter table and the control rule set, comparing predicted values and measured values through simulation and field test, and dynamically updating the hybrid aerodynamic model using an online learning mechanism; S305, updating the fan aerodynamic characteristic prior knowledge base based on the output of the hybrid aerodynamic model, including the aerodynamic parameter table, the control rule set and the abnormal diagnosis library that pass the verification, and generating knowledge base version log record update information; In step S3, the rotational speed and pitch angle dynamic control strategy is established, which comprises: S306, defining a control target based on the joint state vector and the fan aerodynamic characteristic prior knowledge base, the control target including maximizing wind energy capture efficiency, minimizing mechanical load fluctuation and satisfying the thrust coefficient safety boundary; S307, constructing a strategy network based on the joint state vector, and obtaining a continuous rotational speed and pitch angle adjustment instruction satisfying the action space constraint using the strategy network; S308, dynamically correcting the original continuous rotational speed and pitch angle adjustment instruction output by the strategy network based on the aerodynamic parameter table and the control rule set in the prior knowledge base and the control target, the correction including circulation decay compensation and turbulence intensity adjustment; S309, generating a continuous action instruction executable by the fan controller based on the corrected continuous rotational speed and pitch angle adjustment instruction, the continuous action instruction including a time series of rotational speed adjustment amount and pitch angle adjustment amount; S4, optimizing the continuous action instruction based on the boiler operating condition parameters and using a non-dominated sorting genetic algorithm to generate an optimized control solution set; S5, issuing the optimized control solution set to the fan controller, and using the fan controller to execute the optimized control solution set to regulate the speed of the fan. S6, the feedback optimization loop is established, the wind volume, energy consumption and vibration data after execution are input into the Bayesian optimization framework, historical data and real-time data are combined, and a fan control strategy parameter is dynamically adjusted; the Bayesian optimization framework uses a Gaussian process model as a prior distribution, selects a next evaluation point through function collection, iteratively updates model parameters, and stops until a convergence condition or a preset iteration number is reached.

2. The method of claim 1, wherein the method further comprises: The expression of the joint state vector obtained in step S2 is: wherein, denotes a time-frequency feature of a vibration signal, denotes a scale parameter of a wavelet transform, denotes a translation parameter of a wavelet transform, denotes a raw vibration signal collected by a three-axis acceleration sensor, denotes a conjugate, denotes a statistical feature vector of wind speed data, denotes a mean, denotes wind speed data collected by a lidar and an ultrasonic anemometer, denotes a standard deviation, denotes taking a maximum value, denotes taking a minimum value, denotes a statistical feature vector of rotational speed data, denotes fan rotational speed data measured by an encoder, denotes a multi-modal feature fusion vector, denotes a mean vector of a training data set, denotes a standard deviation vector of a training data set.

3. The method of claim 1, wherein the method further comprises: The expression of the dynamically updated hybrid aerodynamic model in step S304 is: wherein, represents the updated blade circulation, represents the initial circulation value, represents the circulation decay rate over time, represents the discretized time step, represents the learning rate, represents the total number of operating points, represents the CFD flow field velocity vector of the th operating point, represents the local time integration variable, represents the BEM flow field velocity vector of the th operating point, represents the updated turbulence intensity, represents the initial turbulence intensity, represents the turbulence intensity adjustment coefficient, represents the number of experimental operating points, represents the experimental lift coefficient of the th operating point, represents the simulated lift coefficient of the th operating point, represents the initial decay rate, represents the decay rate acceleration coefficient, represents the cumulative running time, represents the fan blade design life.

4. The method of claim 1, wherein the method further comprises: The expression of the dynamically corrected original continuous rotating speed and pitch angle adjustment instruction output by the strategy network in step S308 is: Circulation decay compensation term: wherein, denotes the speed adjustment amount after the compensation of the circulation decay, denotes the original speed adjustment amount output by the policy network, denotes the circulation compensation coefficient, denotes the wind speed and the pitch angle corresponding circulation decay rate, denotes the compensation coefficient reference value below the rated wind speed, denotes the compensation decay rate after the wind speed exceeds the rated wind speed, denotes the rated wind speed of the fan; Turbulence intensity adjustment term: wherein, denotes the adjusted pitch angle adjustment amount, denotes the original pitch angle adjustment amount output by the policy network, denotes the turbulence adjustment coefficient, denotes the wind speed and the pitch angle corresponding experimental turbulence intensity, denotes the wind speed and the pitch angle corresponding simulated turbulence intensity, denotes the pitch angle sensitivity coefficient, denotes the optimal pitch angle.

5. The method of claim 1, wherein, In step S4, the optimized control solution set is generated, including: S401, based on the boiler operating condition parameters and the continuous action instruction of the fan, a multi-objective optimization problem is constructed, the optimization problem takes maximizing wind energy efficiency and minimizing mechanical load as core targets, and integrates thrust coefficient safety boundary constraints; S402, based on the multi-objective optimization problem, a non-dominated sorting genetic algorithm is used, and the continuous action instruction is fused when initializing a population; S403, based on the boiler operating condition parameters and the aerodynamic parameter table, a fitness function is defined, the fitness function is used to dynamically adjust the wind energy capture efficiency target weight, an exponential penalty is applied to individuals that violate the thrust coefficient constraint, and a fitness evaluation system that meets the operating condition demand is generated; S404, based on the initialized population and the fitness function, genetic operations are performed for iterative optimization, high-quality individuals are selected through non-dominated sorting, offspring are generated through simulated binary crossover and polynomial mutation, and the maximum iteration number is reached until the iteration optimization result is obtained; S405, based on the boiler operating condition parameters and the iteration optimization result, a non-dominated sorted control solution set is output, and the action instruction sequence that best matches the current load and coal calorific value is selected to generate the optimized control solution set.

6. The method of claim 5, wherein, The expression of the fitness function in step S402 is: wherein, represents a fitness function, represents an individual's corresponding action command vector, represents a wind energy efficiency dynamic weight, represents a wind energy capture efficiency, represents a mechanical load dynamic weight, represents a mechanical load fluctuation, represents a penalty term coefficient, represents a thrust coefficient under the current action command.

7. A kind of intelligent speed regulation system of thermal power plant fan, it is characterized in that, The intelligent speed regulation method of the thermal power plant fan includes the system further comprising: A data acquisition module is configured to acquire multi-modal data of the fan operation and operating condition parameters of the boiler. A preprocessing module is connected with the data acquisition module and configured to preprocess the multi-modal data of the fan operation and the operating condition parameters of the boiler and extract effective feature information. A feature extraction module is connected with the preprocessing module and configured to extract a feature vector reflecting the aerodynamic characteristics and operating state of the fan based on the preprocessed data. A control strategy module is connected with the feature extraction module and configured to generate rotating speed and pitch angle adjustment instructions of the fan according to the extracted feature vector and the intelligent speed regulation method of the thermal power plant fan. An execution module is connected with the control strategy module and configured to receive and execute the rotating speed and pitch angle adjustment instructions to adjust the operating state of the fan. A feedback optimization module is connected with the execution module and the data acquisition module and configured to acquire executed fan operation data, establish a feedback optimization loop, and dynamically adjust the fan control strategy parameter to continuously optimize the fan operation performance.

Images