A wind turbine power curve optimization method, system, medium and product

Abstract

The application provides a wind turbine power curve optimization method, system, medium and product, the method comprising: S1, obtaining the operation data of the SCADA system and performing working condition partition processing to obtain a data subset containing a maximum power tracking area; S2, constructing a second-order polynomial response surface model between wind speed, pitch angle, air density and wind turbine power based on the data subset, fitting and verifying by using a least square method with regularization until the model accuracy reaches a preset threshold; S3, taking the maximization of the model output value as the target, solving the optimal pitch angle within the mechanical constraint range of the pitch angle, generating an optimized control table and embedding the wind turbine controller to realize real-time table lookup control; S4, regularly updating the SCADA data and repeating the above steps to realize continuous optimization of the model and the control strategy. The application aims to realize rapid calibration of the power curve and optimization of dynamic control parameters.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, and in particular to a method, system, medium, and product for optimizing the power curve of a wind turbine. Background Technology

[0002] With the large-scale development and popularization of wind power, the accurate modeling and optimization of wind turbine power curves has become a core technical issue for improving wind energy utilization efficiency and reducing the cost per kilowatt-hour. As a key indicator characterizing wind turbine performance, the power curve directly affects the accuracy of power generation prediction, the formulation of unit control strategies, and the economic benefit assessment of wind power projects.

[0003] Currently, traditional power curve modeling methods mainly rely on the testing and modeling methods recommended by the international standard IEC 61400-12-1, typically based on linear interpolation or piecewise statistical regression techniques. While these methods are simple in structure and easy to implement, they struggle to accurately characterize the complex nonlinear dynamic characteristics under the coupled effects of multiple variables such as wind speed, turbulence intensity, air density, ambient temperature and humidity, and pitch angle. Especially when turbulence intensity is high or environmental conditions change drastically, the response relationship between wind speed and power deviates significantly from the linear assumption, leading to substantial errors in power prediction, typically exceeding 8%. This deviation not only affects the accuracy of power generation forecasts but also limits the real-time optimization capabilities of key control parameters such as pitch and yaw, thus preventing the full realization of power generation potential.

[0004] In recent years, with the development of artificial intelligence technology, machine learning-based methods (such as neural networks, support vector machines, and random forests) have been introduced into power curve modeling, improving prediction accuracy to some extent. However, these "black box" models have significant bottlenecks in engineering applications: First, the models lack interpretability and cannot provide explicit mathematical relationships between variables such as wind speed and turbulence and power output, which is detrimental to unit fault diagnosis, control logic verification, and performance certification that meets standards. Second, the model training and inference processes are computationally complex and resource-intensive, making it difficult to embed into embedded platforms with limited computing power, such as wind turbine main control systems, and failing to meet the millisecond-level response requirements of real-time control. Third, most machine learning models lack online updating and adaptive adjustment mechanisms, resulting in decreased model applicability when facing changes in the wind farm environment or unit performance degradation.

[0005] Furthermore, existing power curve testing and modeling methods are generally based on steady-state wind condition assumptions, failing to fully consider the dynamic wind characteristics present in actual operation, such as instantaneous wind speed fluctuations, turbulent gusts, and rapid changes in wind direction. This leads to significant response lag and adaptability issues in traditional models under dynamic operating conditions: on the one hand, pitch and yaw control cannot be adjusted in a timely manner according to changes in wind conditions, resulting in annual power generation losses; on the other hand, the mechanical loads on the unit under transient wind loads (such as tower vibration and blade fatigue) cannot be effectively suppressed, which not only affects power generation stability but also accelerates structural damage and shortens the turbine's lifespan. Summary of the Invention

[0006] The technical problem to be solved by this invention is: in view of the technical problems existing in the prior art, this invention provides a method, system, medium and product for optimizing the power curve of wind turbine units, aiming to achieve rapid calibration of the power curve and optimization of dynamic control parameters, and provide key technical support for improving the overall power generation efficiency and operation economy of wind farms.

[0007] To solve the above-mentioned technical problems, the technical solution proposed by this invention is as follows:

[0008] A method for optimizing the power curve of a wind turbine includes the following steps:

[0009] Step S1: Obtain the operating data of the wind turbine data acquisition and monitoring system, and perform operating condition partitioning processing on the operating data to obtain data subsets of different partitions. The operating data includes time-series data of wind speed, blade pitch angle, air density and actual power.

[0010] Step S2: Based on the data subsets of the different partitions, construct a second-order polynomial response surface model with wind speed, pitch angle, and air density as input variables and wind turbine power as output variable; use regularized least squares method to fit and verify the model. If the accuracy of the second-order polynomial response surface model does not reach the preset threshold, adjust the regularization parameter and refit and verify the model; if the model accuracy reaches the preset threshold, proceed to step S3.

[0011] Step S3: Using the output value of the second-order polynomial response surface model as the objective function, solve for the optimal pitch angle within the mechanical constraint range of the pitch angle, generate an optimized control table with wind speed and air density as inputs and the optimal pitch angle as output, and embed the optimized control table into the wind turbine controller to implement real-time table lookup control;

[0012] Step S4: Periodically acquire new operating data from the wind turbine data acquisition and monitoring system, and repeat steps S1 to S3 to update the second-order polynomial response surface model and the optimization control table.

[0013] As a further improvement to the method of the present invention: Step S1 further includes performing anomaly cleaning on the operating data, standardizing the wind speed, blade pitch angle and air density data, correcting the actual power to the power value under standard air density, and dividing the operating data into the start-up zone, the maximum power tracking zone and the constant power zone according to the wind speed range, and selecting the data of the maximum power tracking zone for model construction.

[0014] As a further improvement to the method of the present invention: the actual power is corrected to the power value under standard air density using the following formula:

[0015]

[0016] in, To standardize power, This represents the actual measured power of the wind turbine. Standard air density, This is the actual measured air density.

[0017] As a further improvement to the method of the present invention: in step S2, the functional expression of the second-order polynomial response surface model is:

[0018]

[0019] in, The predicted wind turbine power output by the model. v is the wind speed, β is the propeller pitch angle. air density, The coefficients to be fitted are... For random error, For constant terms or intercepts, Let be the coefficients of the quadratic term to be fitted. represents the cross term coefficients to be fitted.

[0020] As a further improvement to the method of the present invention: in step S2, the method of using regularized least squares for model fitting includes:

[0021] Step S201: Transform the second-order polynomial response surface model into a matrix. Where y is the measured power vector obtained based on SCADA data, and X is the design matrix constructed based on wind speed, pitch angle, and air density. Let be the coefficient vector of the model to be solved. This is random error;

[0022] Step S202: Construct a loss function that includes an L2 regularization penalty term:

[0023]

[0024] in, Let λ be the sum of squared residuals, and λ be the regularization strength parameter. Squaring the L2 norm of all coefficient vectors;

[0025] Step S203: Select the λ value that minimizes the root mean square error of the validation set from the preset candidate value range using the cross-validation method, so as to determine the final regularization strength parameter.

[0026] As a further improvement to the method of the present invention: In step S3, the step of solving for the optimal pitch angle within the mechanical constraint range of the pitch angle is to use a sequential quadratic programming algorithm to solve for the optimal pitch angle value within the mechanical constraint range of the pitch angle, with the goal of maximizing the power value output by the model, given a combination of wind speed and air density.

[0027] The present invention also provides a wind turbine power curve optimization system based on a polynomial response surface model for implementing the aforementioned wind turbine power curve optimization method, comprising:

[0028] The data preprocessing module is used to perform runtime data acquisition and operating condition zoning;

[0029] The model building and training module is used to build, fit, and validate the polynomial response surface model.

[0030] The optimization and control table generation module is used to solve for the optimal pitch angle and generate an optimization control table.

[0031] The iterative update module is used to periodically trigger the update process of the model and control table.

[0032] The present invention also provides a wind turbine main control system, which integrates the wind turbine power curve optimization system based on the polynomial response surface model. The wind turbine main control system can load and execute the optimization control table, query the corresponding optimal pitch angle based on the real-time measured wind speed and air density, and control the wind turbine actuator to operate.

[0033] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the wind turbine power curve optimization method described above.

[0034] The present invention also provides a computer program product, including a computer program / instructions, which are programmed or configured to execute the wind turbine power curve optimization method by a processor.

[0035] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0036] 1. This invention achieves high-precision, interpretable nonlinear modeling of wind turbine power characteristics by constructing a second-order polynomial response surface model and combining it with regularized fitting verification. It effectively solves the problems of excessive prediction deviation of traditional linear or piecewise models under complex wind conditions, as well as the poor engineering practicality and difficulty in integrating machine learning black-box models into embedded controllers. It ensures that the model has the ability to approximate complex nonlinear relationships, while also possessing the clear structure and low computational complexity of a white-box model, laying a reliable foundation for real-time optimization control.

[0037] 2. This invention establishes a constrained optimization problem with the goal of maximizing model output and generates an optimized control table, achieving millisecond-level real-time and precise optimization of the pitch angle. It solves the problems of traditional control strategies being unable to dynamically adapt to turbulent changes due to reliance on steady-state power curves, and the annual power generation loss caused by fixed control parameters. Under the premise of ensuring that the pitch angle is always within the mechanical safety constraint range, it maximizes wind energy capture efficiency, thereby simultaneously improving power generation efficiency and unit operation safety.

[0038] 3. By setting a periodic update mechanism, this invention realizes fully automatic iterative calibration of the power curve model and the optimized control strategy. It solves the common industry problem of insufficient long-term applicability of static models and gradual decline in power generation performance caused by factors such as seasonal changes in the environment and drift in unit performance. This enables wind turbines to continuously adapt to their operating environment and their own state, ensuring the long-term effectiveness and robustness of the optimization effect, and providing key technical support for intelligent operation and maintenance of wind farms throughout their entire life cycle. Attached Figure Description

[0039] Figure 1 This is a flowchart of the wind turbine power curve optimization method in an embodiment of the present invention. Detailed Implementation

[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0041] like Figure 1 As shown in the figure, this embodiment provides a method for optimizing the power curve of a wind turbine, including the following steps:

[0042] Step S1: Obtain the operating data of the wind turbine data acquisition and monitoring system, and perform operating condition partitioning processing on the operating data to obtain data subsets of different partitions. The operating data includes time-series data of wind speed, blade pitch angle, air density and actual power.

[0043] Step S2: Based on the data subsets of the different partitions, construct a second-order polynomial response surface model with wind speed, blade pitch angle, and air density as input variables and wind turbine power as output variable; use regularized least squares method to fit and verify the model. If the model accuracy does not reach the preset threshold, adjust the regularization parameter and refit and verify the model; if the model accuracy reaches the preset threshold, proceed to step S3.

[0044] Step S3: With maximizing the output value of the model as the objective function, solve for the optimal pitch angle within the mechanical constraint range of the pitch angle, generate an optimized control table with wind speed and air density as inputs and the optimal pitch angle as output, and embed the optimized control table into the wind turbine controller to implement real-time table lookup control;

[0045] Step S4: Periodically acquire new operating data from the wind turbine data acquisition and monitoring system, and repeat steps S1 to S3 to update the polynomial response surface model and optimize the control table.

[0046] In this embodiment, by dividing the operating data into different operating condition intervals, a second-order polynomial model is constructed with wind speed, pitch angle, and air density as inputs. Regularized least squares method is used for fitting and verification, and the optimal pitch angle is solved to generate a control table which is embedded in the controller. The model parameters are periodically updated to maintain the optimization effect. Compared with existing technologies, traditional linear models only consider the linear relationship between variables and cannot accurately describe the power mutation characteristics under turbulent conditions. This embodiment captures nonlinear effects through second-order polynomial terms, reducing the model prediction error to within 5%. Compared with black-box models such as neural networks, the explicit mathematical expressions provided in this embodiment can be directly used for control logic verification, and the computation time is reduced by about 80%, meeting the real-time requirements of embedded systems. Furthermore, existing static control tables cannot adapt to seasonal changes in air density; the quarterly update mechanism of this scheme ensures that power generation efficiency remains stable throughout the year.

[0047] This embodiment achieves accurate modeling and dynamic optimization of wind turbine power curves. The explicit polynomial model ensures prediction accuracy while providing interpretable variable relationships, facilitating engineers' analysis of control strategies. Regularization methods effectively balance model complexity and generalization ability, avoiding control instability caused by overfitting. Optimized control tables combined with a real-time lookup mechanism reduce pitch angle adjustment response time to milliseconds, significantly improving wind energy capture efficiency. Periodic model updates ensure the control strategy continuously adapts to environmental changes, maintaining optimal power generation under varying climatic conditions throughout the year.

[0048] In step S1 of this embodiment, the abnormal cleaning of the operating data is further performed, and the wind speed, pitch angle and air density data are standardized to correct the actual power to the power value under the standard air density. The operating data is divided into the start-up zone, the maximum power tracking zone and the constant power zone according to the wind speed range, and the data of the maximum power tracking zone is selected for model construction.

[0049] In a specific application embodiment, average data with a 10-minute cycle is acquired from the wind turbine data acquisition and monitoring system (SCADA). The data fields include: wind speed v (m / s), blade pitch angle β (°), air density ρ (kg / m³), and actual power P (kW). After acquiring the raw operating data from the SCADA system, the data is first cleaned to ensure its basic reliability. Raw SCADA data often contains invalid values ​​due to sensor malfunctions or transmission errors. This embodiment employs a two-stage cleaning strategy: first, amplitude limiting filtering, which removes data that clearly exceeds reasonable ranges based on physical principles, such as records with wind speeds less than zero (v<0) or actual power abnormally exceeding 120% of rated power (P>1.2Pr); second, outlier removal, which aims to eliminate statistically significant outliers, such as removing shutdown status data with continuously zero power (P = 0) and other abnormal fluctuation data identified through statistical methods (such as box plots).

[0050] Secondly, based on data cleaning, the key input variables are standardized. In this embodiment, wind speed (v), blade pitch angle (β), and air density (ρ) are standardized using Z-score, with the specific formula as follows:

[0051] (1)

[0052] in, For the standardized variable values, x For original variables, μ The sample mean. σ This represents the sample standard deviation.

[0053] In this embodiment, the actual power is corrected to the standard air density using the following formula ( Power value (=1.225kg / m³):

[0054] (2)

[0055] in, To standardize power, This represents the actual measured power of the wind turbine. Standard air density, This is the actual measured air density.

[0056] To construct an optimization model focusing on energy capture efficiency, in this embodiment, the operation data is divided into three typical working condition zones according to the wind speed range:

[0057] Startup zone (0 < v < Vin): When the wind speed is lower than the cut-in wind speed, the wind turbine does not generate electricity or is about to start. The data in this zone has no value for power generation analysis.

[0058] Maximum power tracking zone (Vin < v < Vr): When the wind speed is between the cut-in wind speed and the rated wind speed, the control objective of the wind turbine is to track the maximum wind energy utilization coefficient (Cp_max) to capture as much wind energy as possible. The data in this zone directly reflects the core relationship between wind energy capture and control parameters, and is the necessary data basis for constructing the power curve optimization model in this embodiment.

[0059] Constant power zone (Vr < v < Vout): When the wind speed is higher than the rated wind speed, the wind turbine maintains the power at the rated value (Pr) by adjusting the pitch angle and other operations. The data in this stage reflects the power limitation control strategy.

[0060] In this embodiment, through a systematic preprocessing process, a high-quality data subset that has been cleaned, standardized, and corrected for air density in the "maximum power tracking zone" is finally selected from the original data for the subsequent construction of the second-order polynomial response surface model. This rigorous data preparation process is the key prerequisite for ensuring that the model has high accuracy and strong generalization ability.

[0061] In step S2 of this embodiment, the functional expression of the second-order polynomial response surface model is:

[0062] (3)

[0063] Where, is the predicted wind turbine power output by the model, , v is the wind speed, β is the pitch angle, is the air density, are the coefficients to be fitted, is the random error, is the constant term or intercept, are the quadratic coefficients to be fitted, are the cross-term coefficients to be fitted.

[0064] In this embodiment, the method of using the least squares method with regularization for model fitting includes:

[0065] Step S201: Convert the second-order polynomial response surface model into a matrix:

[0066] ;

[0067] Where y is the measured power vector obtained based on SCADA data, and X is the design matrix constructed based on wind speed, pitch angle, and air density. Let be the coefficient vector of the model to be solved. This is random error;

[0068] Step S202: Construct a loss function that includes an L2 regularization penalty term:

[0069] (5)

[0070] in, Let be the sum of squared residuals, and λ be the regularization strength parameter, a hyperparameter, and λ≥0, controlling the strength of the penalty: λ=0 degenerates into ordinary least squares. The square of the L2 norm of all coefficient vectors (usually excluding the penalized intercept term). ).

[0071] Step S203: Select the λ value that minimizes the root mean square error of the validation set from the preset candidate value range using the cross-validation method, so as to determine the final regularization strength parameter.

[0072] Specifically, the steps for model fitting using regularized least squares method include:

[0073] Step 1: Transform the problem into matrix form

[0074] (1) Constructing model equations:

[0075] (6)

[0076] (2) Construct the design matrix (X) and predicted power (y):

[0077] Suppose there are m sets of SCADA data samples. For each set (v, β, ρ, P), convert it into the feature form of the model.

[0078] ①Predicted power y:

[0079] (7)

[0080] in, The measured power of the extension unit in the m-th data sample group is given.

[0081] ② Design matrix X (containing constant terms, linear terms, square terms, and cross terms):

[0082] (8)

[0083] ③ Coefficient vector θ:

[0084] (9)

[0085] Therefore, the model equation is changed to formula (4).

[0086] Step 2: Define the loss function with L2 regularization

[0087] (1) The loss function of standard least squares is the residual sum of squares (RSS):

[0088] (10)

[0089] L2 regularization adds a penalty term to this, which is the sum of squares of all coefficients (excluding the intercept). ) times λ;

[0090] (2) Define the loss function J(θ) as shown in formula (5):

[0091] Step 3: Solve for the optimal coefficient vector

[0092] (1) Differentiate the loss function: Differentiate J(θ) with respect to θ and set it equal to 0:

[0093] (11)

[0094] It is an identity matrix, but the corresponding intercept term is 0, indicating that the intercept term is not penalized.

[0095] (2) The normal equation is obtained:

[0096] (12)

[0097] (2) Solve :

[0098] The optimal coefficients can be obtained by directly solving the system of linear equations:

[0099] (13)

[0100] Step 4: Selection of hyperparameter λ

[0101] The value of λ needs to be determined through cross-validation of the data, specifically including:

[0102] (1) Prepare data: Divide the dataset into a training set (for fitting the model) and a validation set (for evaluating the model performance).

[0103] (2) Define the λ grid: Select a range of values ​​for λ, for example [0.01, 0.1, 1.0, 10.0, 100.0].

[0104] (3) Cross-validation cycle:

[0105] ① For each candidate λ value, use the training set to solve according to step 3. .

[0106] ② Use the obtained model to predict the data of the validation set and calculate performance metrics (such as root mean square error RMSE).

[0107] (4) Select the λ value that minimizes RMSE (or other metrics) on the validation set.

[0108] (5) Retrain the final model using the optimal λ value and all training data.

[0109] After obtaining the final model, this embodiment requires rigorous accuracy verification to determine whether it meets the conditions for entering the optimization control stage. The verification criteria include:

[0110] Goodness of fit: The model should adequately explain the variability of the data, and the coefficient of determination should satisfy R2 > 0.95.

[0111] Prediction error: The prediction accuracy of the model must meet the engineering requirements, and its relative root mean square error must meet RMSE < 5% * Pr (where Pr is the rated power of the wind turbine).

[0112] Finally, an iterative optimization mechanism is established to ensure the robustness of the solution. If the final model simultaneously meets both of the above accuracy acceptance criteria, the model is deemed qualified; otherwise, the regularization parameters are adjusted.

[0113] In step S3 of this embodiment, solving for the optimal pitch angle within the mechanical constraint range of the pitch angle involves using a sequential quadratic programming algorithm to obtain the optimal pitch angle value within the mechanical constraint range of the pitch angle, with the goal of maximizing the power output value of the model for a given combination of wind speed and air density.

[0114] The specific power curve optimization and control implementation process is as follows:

[0115] Step 1: Construct a constrained nonlinear optimization problem. The objective function is: Constraints: .

[0116] Step 2: For each wind speed and air density ( v , ρ The optimal pitch angle is obtained by combining the following methods and using a sequential quadratic programming algorithm to solve the constrained optimization problem. :

[0117] =argmax P model(v , β , ρ ).

[0118] Step 3: Compile the obtained optimal pitch angle results into an optimization control lookup table, the format of which is shown in Table 1:

[0119] Table 1 Optimization Control Table

[0120]

[0121] Step 4: Write the generated optimized control table into the fan PLC / SCADA system and implement a real-time lookup control strategy: based on the real-time measured wind speed. v and air density ρ Query the control table and adjust the pitch angle to the optimal value. .

[0122] Step 5: Establish a continuous optimization mechanism for the model:

[0123] Quarterly Data Update: Obtain new SCADA data each quarter, then return to step 1 to retrain the model.

[0124] Update cycle: A 90-day update cycle is recommended to adapt to fluctuations in environmental parameters caused by seasonal changes.

[0125] Automatic control table update: Automatically update the model and generate a new optimized control table to ensure that the control strategy remains optimal.

[0126] Through the above technical solutions, this embodiment achieves accurate modeling and dynamic optimization of the wind turbine power curve. The explicit polynomial model, while ensuring prediction accuracy, provides interpretable variable relationships, facilitating engineers' analysis of control strategies. Regularization methods effectively balance model complexity and generalization ability, avoiding control instability caused by overfitting. Optimized control tables combined with a real-time lookup mechanism reduce the pitch angle adjustment response time to milliseconds, significantly improving wind energy capture efficiency. Periodic model updates ensure that the control strategy continuously adapts to environmental changes, maintaining optimal power generation under various climatic conditions throughout the year.

[0127] This embodiment solves the technical bottlenecks of traditional methods in nonlinear modeling, dynamic adaptability, and online calibration, and provides an effective technical solution for intelligent operation and maintenance and performance improvement of wind farms.

[0128] This embodiment also provides a wind turbine power curve optimization system based on a polynomial response surface model for implementing a wind turbine power curve optimization method, including:

[0129] The data preprocessing module is used to perform runtime data acquisition and operating condition zoning;

[0130] The model building and training module is used to build, fit, and validate the polynomial response surface model.

[0131] The optimization and control table generation module is used to solve for the optimal pitch angle and generate an optimization control table.

[0132] The iterative update module is used to periodically trigger the update process of the model and control table.

[0133] The data preprocessing module is responsible for collecting, cleaning, and zoning wind turbine operating data. Specifically, it uses amplitude-limiting filtering algorithms to remove outlier data and divides the system into start-up, maximum power tracking, and constant power zones based on wind speed ranges. Standardization eliminates dimensional differences, providing high-quality input for model building. The model building and training module performs parameter fitting and verification based on a second-order polynomial response surface model. It uses L2 regularized least squares to solve for coefficients and combines cross-validation to select the regularization strength parameter, avoiding overfitting and ensuring the model's generalization ability in embedded devices. The optimization and control table generation module solves for the optimal pitch angle and generates a control table. It uses a sequential quadratic programming algorithm to perform global optimization within the mechanical constraints of the pitch angle, generating an optimal pitch angle lookup table with wind speed and air density as inputs. The iterative update module periodically triggers updates to the model and control table. It uses timers or data volume threshold triggering mechanisms to re-execute the data collection, model training, and optimization processes, adapting to environmental parameter fluctuations caused by seasonal changes.

[0134] Specifically, the data preprocessing module first obtains raw operational data from the SCADA system, generates a standardized dataset through outlier filtering and air density correction, and divides the operating conditions according to wind speed ranges. The model building and training module constructs a second-order polynomial model based on the preprocessed data, fits the coefficients using regularized least squares, verifies the model accuracy, and outputs a qualified model. The optimization and control table generation module solves for the optimal blade pitch angle for different wind speed and air density combinations, forms a two-dimensional lookup table, and embeds it into the wind turbine controller. The iterative update module periodically collects new data, re-executes the preprocessing, modeling, and optimization processes, and achieves dynamic updates of the model and control table.

[0135] Compared to existing technologies, traditional systems typically employ static control tables and lack model update mechanisms, making them unable to adapt to dynamic wind conditions and environmental changes. Existing machine learning-based optimization systems suffer from problems such as uninterpretable models and high computational resource consumption, making them difficult to run in real-time on embedded controllers. This system, through an explicit multinomial model and regularized fitting methods, reduces computational complexity while maintaining model accuracy, and continuously optimizes the control strategy by combining it with a periodic update mechanism.

[0136] This embodiment also provides a wind turbine main control system, which integrates a wind turbine power curve optimization system based on a polynomial response surface model. The wind turbine main control system can load and execute the optimization control table, query the corresponding optimal pitch angle based on the real-time measured wind speed and air density, and control the wind turbine actuator to operate.

[0137] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements a method for optimizing the power curve of a wind turbine.

[0138] This embodiment also provides a computer program product, including a computer program / instruction, which is programmed or configured to execute a wind turbine power curve optimization method via a processor.

[0139] Those skilled in the art will understand that the above embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the functions specified in one or more boxes. The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

[0140] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Therefore, any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should fall within the protection scope of the present invention.

Claims

1. A method for optimizing the power curve of a wind turbine generator set, characterized in that, Includes the following steps: Step S1: Obtain the operating data of the wind turbine data acquisition and monitoring system, and perform operating condition partitioning processing on the operating data to obtain data subsets of different partitions. The operating data includes time-series data of wind speed, blade pitch angle, air density and actual power. Step S2: Based on the data subsets of the different partitions, construct a second-order polynomial response surface model with wind speed, blade pitch angle, and air density as input variables and wind turbine power as output variable; use regularized least squares method to fit and verify the model. If the model accuracy does not reach the preset threshold, adjust the regularization parameter and refit and verify the model; if the model accuracy reaches the preset threshold, proceed to step S3. Step S3: Using the output value of the second-order polynomial response surface model as the objective function, solve for the optimal pitch angle within the mechanical constraint range of the pitch angle, generate an optimized control table with wind speed and air density as inputs and the optimal pitch angle as output, and embed the optimized control table into the wind turbine controller to implement real-time table lookup control; Step S4: Periodically acquire new operating data from the wind turbine data acquisition and monitoring system, and repeat steps S1 to S3 to update the second-order polynomial response surface model and the optimization control table; In step S1, the process further includes cleaning the operational data to remove anomalies, standardizing the wind speed, pitch angle, and air density data, correcting the actual power to the power value under standard air density, dividing the operational data into a start-up zone, a maximum power tracking zone, and a constant power zone according to the wind speed range, and selecting the data from the maximum power tracking zone for model construction. In step S2, the method of using regularized least squares for model fitting includes: Step S201: Transform the second-order polynomial response surface model into a matrix. Where y is the measured power vector obtained based on SCADA data, and X is the design matrix constructed based on wind speed, pitch angle, and air density. Let be the coefficient vector of the model to be solved. This is random error; Step S202: Construct a loss function that includes an L2 regularization penalty term: in, Let λ be the sum of squared residuals, and λ be the regularization strength parameter. Squaring the L2 norm of all coefficient vectors; Step S203: Select the λ value that minimizes the root mean square error of the validation set from the preset candidate value range using the cross-validation method, so as to determine the final regularization strength parameter.

2. The wind turbine power curve optimization method according to claim 1, characterized in that, The actual power is corrected to the power value under standard air density using the following formula: in, To standardize power, This represents the actual measured power of the wind turbine. Standard air density, This is the actual measured air density.

3. The wind turbine power curve optimization method according to claim 1, characterized in that, In step S2, the functional expression of the second-order polynomial response surface model is: in, The predicted wind turbine power output by the model. v is the wind speed, β is the propeller pitch angle. air density, The coefficients to be fitted are... For random error, For constant terms or intercepts, The coefficients of the quadratic term to be fitted are... represents the cross term coefficients to be fitted.

4. The wind turbine power curve optimization method according to claim 1, characterized in that, In step S3, the step of solving for the optimal pitch angle within the mechanical constraint of the pitch angle is to take a given combination of wind speed and air density, and aim to maximize the power value output by the second-order polynomial response surface model. Within the mechanical constraint of the pitch angle, the optimal pitch angle value is obtained by using a sequential quadratic programming algorithm.

5. A wind turbine power curve optimization system based on a polynomial response surface model for implementing the method described in any one of claims 1 to 4, characterized in that, include: The data preprocessing module is used to perform runtime data acquisition and operating condition zoning; The model building and training module is used to build, fit, and validate the second-order polynomial response surface model. The optimization and control table generation module is used to solve for the optimal pitch angle and generate an optimization control table. The iterative update module is used to periodically trigger the update process of the model and control table.

6. A wind turbine main control system, characterized in that, The wind turbine main control system integrates the wind turbine power curve optimization system based on the polynomial response surface model as described in claim 5. The wind turbine main control system can load and execute the optimization control table, query the corresponding optimal pitch angle based on the real-time measured wind speed and air density, and control the wind turbine actuator to operate.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the wind turbine power curve optimization method as described in any one of claims 1 to 4.

8. A computer program product, comprising a computer program, characterized in that, The computer program is programmed or configured to execute the wind turbine power curve optimization method according to any one of claims 1 to 4 via a processor.

Images