On-line self-correction method and system for wake model parameters based on DE-SQP combined framework
By employing an online self-calibration method for wake model parameters within the DE-SQP joint framework, and utilizing differential evolution and sequential quadratic programming algorithms to optimize wake model parameters, the problem of insufficient generalization ability of the double Gaussian wake model is solved, achieving high-precision wind farm wake prediction and improved power generation performance.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 广东粤电阳江海上风电有限公司
- Filing Date
- 2026-01-21
- Publication Date
- 2026-06-02
AI Technical Summary
The empirical parameters of existing double Gaussian wake models are mostly calibrated based on specific operating conditions, which lacks generalization ability and makes it difficult to adapt to complex and ever-changing atmospheric conditions, resulting in power generation losses in wind farms.
An online self-calibration method for wake model parameters based on the DE-SQP joint framework is adopted. The global search optimization is performed by differential evolution algorithm, and the local fine optimization is performed by sequential quadratic programming algorithm. The wake model parameters are calibrated in real time, the relationship between the equivalent wind speed and thrust coefficient of the wind turbine generator is established, and the multidimensional empirical parameters of the wake model are optimized.
It significantly reduced the sum of squared residuals between the model's predicted power and the measured power, improved the model's prediction accuracy and generalization ability under different atmospheric stability conditions, adapted to different types of wind turbines and wind farm layouts, and reduced prediction errors by more than 96%.
Smart Images
Figure CN122132694A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the technical field of wind power generation, and in particular to an online self-calibration method and system for wake model parameters based on the DE-SQP joint framework. Background Technology
[0002] With the rapid growth of wind power installed capacity, the power generation loss caused by the wake effect of wind farms is becoming increasingly prominent. Wake control is an important means to improve the overall performance of wind farms, but its effectiveness is highly dependent on the prediction accuracy of the wake model. Although existing double Gaussian wake models outperform traditional Jesen models in the near-wake region, their empirical parameters are mostly calibrated based on specific operating conditions, lacking generalization ability and failing to adapt to complex and variable atmospheric conditions. Currently, there is a lack of a systematic multi-scenario parameter calibration scheme to improve the model's prediction robustness across the entire wake region. Summary of the Invention
[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide an online self-calibration method and system for wake model parameters based on the DE-SQP joint framework.
[0004] To achieve the above objectives, the technical solution provided by this invention is as follows:
[0005] An online self-calibration method for wake model parameters based on the DE-SQP joint framework includes:
[0006] Construct a wake calculation system that includes a wind turbine model, a wind speed model, and a wake model;
[0007] The measured power data of the wind turbine generator set under the target operating condition is obtained, and the predicted power data under the corresponding operating condition is calculated using the wind turbine model and wind speed model in the wake calculation system.
[0008] Based on measured power data and predicted power data, an optimization objective function is set;
[0009] Based on the objective function, the differential evolution algorithm is used to perform global search optimization on the multidimensional empirical parameters of the wake model, obtain the global optimal preliminary parameter vector, avoid getting trapped in local optima and provide a good initial optimization point;
[0010] Starting with the initial parameter vector, a sequential quadratic programming algorithm is used for local fine-tuning, and its fast convergence characteristic is used to obtain the final correction parameters.
[0011] The final correction parameters are applied to the wake model in real time to complete the online self-calibration of the model parameters.
[0012] Furthermore, the wind turbine model is a simplified model of the wind turbine generator consisting of a power curve and a thrust coefficient curve. This model is fitted using field measurement data or performance curves provided by the manufacturer to establish the equivalent wind speed at the wind turbine rotor. With predicted power Thrust coefficient The mapping relationship between them is used to evaluate the power generation performance of a single wind turbine and the entire field;
[0013] Among them, the equivalent wind speed at the wind turbine rotor With predicted power The mapping relationship is as follows:
[0014] ;
[0015] This is a power lookup table function;
[0016] Equivalent wind speed at the wind turbine impeller With thrust coefficient The mapping relationship is as follows:
[0017] ;
[0018] This is a lookup function for the thrust coefficient.
[0019] Furthermore, the wind speed model uses the average wind speed at multiple sampling points as the equivalent wind speed at the wind turbine rotor. The sampling points are distributed in concentric circles on the hub section, and the inner and outer radii are set according to the blade length and structural characteristics.
[0020] Equivalent wind speed at the wind turbine impeller The calculation formula is:
[0021]
[0022] in, No. Wind speed collected at each sampling point This represents the number of sampling points.
[0023] Furthermore, the wake model is a double Gaussian wake model, used to predict wake velocity loss. The relationship between the velocity loss at a point in the wake region and its spatial location is as follows:
[0024]
[0025] in, This represents the speed loss value. Downstream distance; Horizontal distance; Vertical distance; The incoming wind speed at the height of the wheel hub; The amplitude coefficient; The extreme radius of the double-peaked wake; The standard deviation of the wake width; It is a natural exponential function; Radial distance:
[0026]
[0027] Wheel hub height;
[0028] Amplitude coefficient From thrust coefficient Hub diameter The calculation formula is determined by both the distance to the downstream side and the distance to the upstream side, as follows:
[0029]
[0030] in, , These are two transition parameters, expressed as follows:
[0031]
[0032]
[0033] in, It is the error function; The standard deviation of the wake width, and its distance from the downstream. The relationship is as follows:
[0034]
[0035] Extreme radius of the double-peaked wake The calculation formula is as follows:
[0036]
[0037] , , The parameters for fitting the wake are calculated using the following formula:
[0038]
[0039] in, For turbulence intensity, , , , , , , , , , , , , , , These are empirical parameters; once these 15 parameters are determined, the wake model can be uniquely determined.
[0040] Furthermore, the root mean square error of the measured power data and the predicted power data is... As an optimization objective:
[0041]
[0042] in, For the first Measured power at each location; For parameter vector based The predicted power; This represents the total number of data points used for correction.
[0043] Furthermore, a differential evolution algorithm is used to globally search and optimize the multidimensional empirical parameters of the wake model, including:
[0044] Generate by An initial population consisting of individuals Each individual in the initial population Each corresponds to a set of parameter vectors. During the mutation phase, for the first... Target individuals in the population Generate mutation vectors using a difference strategy:
[0045]
[0046] in, Three distinct individuals were randomly selected from the initial population. To control the variation factor of the difference amplitude; the crossover operation will change the variation vector. With target individuals Dimensionally blended to generate test vectors For the first test vector Dimensional components:
[0047]
[0048] in, For crossover probability, For random probability, It is a randomly selected dimension index. For the mutation vector The dimensional components, For target individuals The Dimensional components; in the selection operation, by comparing the trial vectors With target individuals The fitness of the population determines which better individuals are retained for the next generation. The fitness function is the objective function. The algorithm terminates and outputs the initial parameter vector that represents the current global optimum when the maximum number of iterations is reached or the fitness change is less than a set threshold. .
[0049] Furthermore, with the initial parameter vector Starting from point , a sequential quadratic programming algorithm is used for local fine-tuning optimization. At the th... In the next iteration, based on the current parameters Constructing QP subproblems:
[0050]
[0051] in, The search direction vector, The BFGS approximation of the Hessian matrix of the objective function. This represents the objective function that needs to be minimized. Represents the gradient operator. The total number of constraints. It is the transpose operator. Given the parameter constraints, solve this QP subproblem to obtain the optimal search method. Then, the step size independent variable is obtained through one-dimensional line search. Determine the first Step size of the next iteration:
[0052]
[0053] Update parameter vector The iterative process terminates when the first-order necessary condition of KKT is met.
[0054]
[0055] in, For the Lagrange multipliers corresponding to the inequality constraints, This is a locally optimal solution; the above process can be used to obtain the correction parameters that satisfy the physical constraints and minimize the prediction error.
[0056] Furthermore, to achieve the above objectives, the present invention also provides an online self-calibration system for wake model parameters based on the DE-SQP joint framework, used to implement the above-mentioned online self-calibration method for wake model parameters based on the DE-SQP joint framework. It includes a wake calculation system construction module, a measured power acquisition module, a predicted power calculation module, an optimization objective function setting module, a global parameter search optimization module, a local parameter optimization module, and an application module.
[0057] in,
[0058] The wake calculation system construction module is used to construct a wake calculation system that includes a wind turbine model, a wind speed model, and a wake model.
[0059] The measured power acquisition module is used to acquire measured power data of the wind turbine generator set under the target operating conditions;
[0060] The predicted power calculation module uses the wind turbine model and wind speed model in the wake calculation system to calculate the predicted power data under the corresponding operating conditions.
[0061] The optimization objective function setting module sets the optimization objective function based on measured power data and predicted power data;
[0062] The parameter global search optimization module uses the differential evolution algorithm to perform global search optimization on the multidimensional empirical parameters of the wake model to obtain the globally optimal preliminary parameter vector.
[0063] The parameter local optimization module uses the initial parameter vector as the starting point and employs a sequential quadratic programming algorithm to perform local fine optimization to obtain the final correction parameters.
[0064] The application module applies the final correction parameters to the wake model in real time.
[0065] Compared with existing technologies, the principles and advantages of this technical solution are as follows:
[0066] 1. A data-driven optimization framework is constructed using the Differential Evolutionary Algorithm (DE), directly utilizing measured power data collected from wind farms as the optimization target input. The DE algorithm employs a population-parallel search mechanism, simultaneously evaluating NP sets of parameter combinations (NP being the population size) in each iteration. Its computational complexity is only approximately O(NP×G) (G being the number of iterations), reducing it by 2-3 orders of magnitude compared to numerical simulation. This "run-and-calibrate" mode overcomes the limitations of traditional offline calibration, enabling the model to respond to changes in atmospheric conditions in real time.
[0067] 2. The dual-Gaussian wake model contains 15 empirical parameters, and traditional fixed parameters cannot adapt to the variable atmospheric stability. This technical solution adopts a collaborative strategy of DE global search + SQP local fine-tuning: DE conducts a broad exploration in the 15-dimensional parameter space through mutation and crossover operations to avoid getting trapped in local optima; the SQP algorithm, based on the high-quality initial points provided by DE, uses the BFGS quasi-Newton method to construct the Hessian matrix approximation of the objective function, and achieves a deep and refined search by solving quadratic programming subproblems. This two-stage optimization from coarse to fine makes the parameter combination closer to the global optimum, directly minimizing the RMSE objective function, thereby reducing the sum of squared residuals between the model's predicted power and the measured power by more than 96%.
[0068] 3. Model the 15 empirical parameters as thrust coefficients. and turbulence intensity The power function relationship conforms to the physical similarity law of turbulent wakes, and is applicable to different impeller diameters. The thrust characteristic curves and wind field layouts have inherent compatibility, and can be quickly adapted to new scenarios with only re-optimization. It has strong generalization ability and can be applied to different types of wind turbines, wake models and wind field layouts.
[0069] 4. By combining global DE search with local SQP fine-tuning, local optima are effectively avoided, thus improving the stability of parameter correction. Attached Figure Description
[0070] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the services required in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0071] Figure 1 This is a flowchart illustrating the principle of the online self-calibration method for wake model parameters based on the DE-SQP joint framework in an embodiment of the present invention.
[0072] Figure 2 This is a connection block diagram of the online self-calibration system for wake model parameters based on the DE-SQP joint framework according to an embodiment of the present invention;
[0073] Figure 3 Wind profile diagrams at the CFD inlet under different atmospheric stability conditions;
[0074] Figure 4 The diagram shows the time-averaged velocity distribution at the wheel hub height horizontal plane under three typical atmospheric stability conditions.
[0075] Figure 5This graph shows the variation of wind turbine power with downstream distance under different atmospheric stability conditions.
[0076] Figure 6 The graph shows a comparison of the predicted power values before and after model correction with the CFD results at locations 5D, 9D, and 13D under different atmospheric stability conditions. Detailed Implementation
[0077] The present invention will be further described below with reference to specific embodiments:
[0078] like Figure 1 As shown in this embodiment, the online self-calibration method for wake model parameters based on the DE-SQP joint framework includes the following steps:
[0079] S1. Construct a wake calculation system that includes a wind turbine model, a wind speed model, and a wake model;
[0080] In this step,
[0081] The wind turbine model simplifies the wind turbine generator into a reduced-order model consisting of a power curve and a thrust coefficient curve. This model is fitted using field measurement data or performance curves provided by the manufacturer to establish the equivalent wind speed at the wind turbine rotor. With predicted power Thrust coefficient The mapping relationship between them is used to evaluate the power generation performance of a single wind turbine and the entire field;
[0082] Among them, the equivalent wind speed at the wind turbine rotor With predicted power The mapping relationship is as follows:
[0083] ;
[0084] This is a power lookup table function;
[0085] Equivalent wind speed at the wind turbine impeller With thrust coefficient The mapping relationship is as follows:
[0086] ;
[0087] This is a lookup function for the thrust coefficient.
[0088] correspond and The lookup results are shown in Table 1:
[0089] Table 1. Power and thrust coefficient prediction mapping table based on equivalent wind speed
[0090]
[0091] The wind speed model uses the average wind speed at 14 sampling points as the equivalent wind speed at the wind turbine rotor. The sampling points are distributed in concentric circles on the hub section. Specifically, the sampling method adopts a ring arrangement, with two rings of sampling points set on the impeller plane: the outer circle has a radius of 3 / 4R, and the inner circle has a radius of 1 / 2R, where R is the impeller radius.
[0092] Equivalent wind speed at the wind turbine impeller The calculation formula is:
[0093]
[0094] in, No. Wind speed collected at each sampling point This represents the number of sampling points.
[0095] The wake model is a double Gaussian wake model, used to predict wake velocity loss. The relationship between the velocity loss at a point in the wake region and its spatial location is as follows:
[0096]
[0097] in, This represents the speed loss value. Downstream distance; Horizontal distance; Vertical distance; The incoming wind speed at the height of the wheel hub; The amplitude coefficient; The extreme radius of the double-peaked wake; The standard deviation of the wake width; It is a natural exponential function; Radial distance:
[0098]
[0099] Wheel hub height;
[0100] Amplitude coefficient From thrust coefficient Hub diameter The calculation formula is determined by both the distance to the downstream side and the distance to the upstream side, as follows:
[0101]
[0102] in, , These are two transition parameters, expressed as follows:
[0103]
[0104]
[0105] in, It is the error function; The standard deviation of the wake width, and its distance from the downstream. The relationship is as follows:
[0106]
[0107] Extreme radius of the double-peaked wake The calculation formula is as follows:
[0108]
[0109] , , The parameters for fitting the wake are calculated using the following formula:
[0110]
[0111] in, For turbulence intensity, , , , , , , , , , , , , , , These are empirical parameters; once these 15 empirical parameters are determined, the wake model can be uniquely determined.
[0112] The specific settings for the initial 15 empirical parameters are shown in Table 2:
[0113] Table 2. Initial Empirical Parameters for the Double Gaussian Wake Model
[0114]
[0115] S2. Obtain the measured power data of the wind turbine generator set under the target operating condition, and use the wind turbine model and wind speed model in the wake calculation system to calculate the predicted power data under the corresponding operating condition.
[0116] In this step,
[0117] The predicted power data is based on a wind field consisting of two fans arranged in series, where the rear fan is only affected by the wake of the front fan. By adjusting the coordinate position of the rear fan and the inflow turbulence intensity, predicted power data under different spacing and turbulence conditions can be obtained.
[0118] The measured power data were obtained from CFD simulations, including three operating conditions: stable atmospheric boundary layer (SBL), neutral atmospheric boundary layer (NBL), and convective boundary layer (CBL). The generation mechanism parameter settings for different atmospheric stability conditions are shown in Table 3. The inflow wind velocity for all three operating conditions was 10 m / s, and the turbulence intensities were 0.1, 0.08, and 0.06, respectively. The inlet mean velocity, temperature, and turbulent kinetic energy profiles are shown in Table 3. Figure 3 As shown. The simulation lasted for 1 hour. The time-averaged velocity field distribution at the hub height horizontal plane and the variation of wind turbine power with downstream distance under different atmospheric stability conditions are shown in the figures. Figure 4 and Figure 5 As shown.
[0119] Table 3 Parameter settings for different atmospheric stability generation mechanisms
[0120]
[0121] To assess the impact of CFD wake on the power generation performance of downstream units, six monitoring sections (3D / 5D / 7D / 9D / 11D / 13D) were arranged at the hub height level, with 18 equidistant sampling points set along the spanwise direction for each section. The theoretical power of the generator was calculated based on the average velocity at each sampling point, and the instantaneous power was time-averaged to obtain the relationship between average power and atmospheric stability and downstream distance. The results are shown in Table 4.
[0122] Table 4. Average 1-hour power generation (kW) of wind turbines at different downstream distances
[0123]
[0124] S3. Based on measured power data and predicted power data, set the optimization objective function;
[0125] Predicting power from wind turbine models at 3D, 7D, and 11D sections of the wake region. With CFD simulation power Minimizing the mean squared root error (RMSE) is the optimization objective. This objective function is used not only in the parameter optimization process but also as a core metric for subsequent model accuracy evaluation. The objective function expression is:
[0126]
[0127] in, These are the 15 parameter vectors to be optimized.
[0128] S4. Based on the objective function, the differential evolution algorithm is used to perform global search optimization on the multidimensional empirical parameters of the wake model, obtain the global optimal preliminary parameter vector, avoid getting trapped in local optima and provide a good initial optimization point;
[0129] In this step, the differential evolution algorithm is used to perform a global search optimization of the multidimensional empirical parameters of the wake model, including:
[0130] Generate by An initial population consisting of individuals Each individual in the initial population Each corresponds to a set of parameter vectors. During the mutation phase, for the first... Target individuals in the population Generate mutation vectors using a difference strategy:
[0131]
[0132] in, , To control the variation factor of the difference amplitude; the crossover operation will change the variation vector. With target individuals Dimensionally blended to generate test vectors For the first test vector Dimensional components:
[0133]
[0134] in, For crossover probability, For random probability, It is a randomly selected dimension index. For the mutation vector The dimensional components, For target individuals The Dimensional components; in the selection operation, by comparing the trial vectors With target individuals The fitness of the population determines which better individuals are retained for the next generation. The fitness function is the objective function. The algorithm terminates and outputs the initial parameter vector that represents the current global optimum when the maximum number of iterations is reached or the fitness change is less than a set threshold. .
[0135] This step calls the `scipy.optimize.differential_evolution` function, using RMSE as the optimal optimization objective. It takes the parameter boundaries shown in Table 5 and the differential evolution algorithm control parameters shown in Table 6 as input conditions, and iterates continuously through mutation, crossover, and selection operations to finally obtain a set of globally optimal preliminary parameter vectors. .
[0136] Table 5. Empirical parameters and boundary conditions for the double Gaussian wake model.
[0137]
[0138] Table 6 DE Algorithm Control Parameters
[0139]
[0140] S5. Starting with the initial parameter vector, continue to use RMSE as the objective function and apply the parameter boundary constraints listed in Table 5. Use the sequential quadratic programming algorithm for local fine optimization and utilize its fast convergence characteristics to obtain the final correction parameters.
[0141] This step is in the first... In the next iteration, based on the current parameters Constructing QP subproblems:
[0142]
[0143] in, The search direction vector, The BFGS approximation of the Hessian matrix of the objective function. This represents the objective function that needs to be minimized. Represents the gradient operator. The total number of constraints. It is the transpose operator. Given the parameter constraints, solve this QP subproblem to obtain the optimal search method. Then, the step size independent variable is obtained through one-dimensional line search. Determine the first Step size of the next iteration:
[0144]
[0145] Update parameter vector The iterative process terminates when the first-order necessary condition of KKT is met.
[0146]
[0147] in, For the Lagrange multipliers corresponding to the inequality constraints, This is a locally optimal solution; the above process can be used to obtain the correction parameters that satisfy the physical constraints and minimize the prediction error.
[0148] S6. Apply the final correction parameters to the wake model in real time to complete the online self-calibration of the model parameters.
[0149] To demonstrate the effectiveness of the method described in this invention, the following verification was performed:
[0150] 1) Training set validation:
[0151] After parameter calibration of the joint optimization framework of Differential Evolution (DE) and Sequential Quadratic Programming (SQP), the optimized model parameters were used to recalculate the wind turbine power at downstream distances of 3D, 7D, and 11D. The calculated results were compared with the measured power obtained based on CFD simulation, and the root mean square error (RMSE) was used as the core evaluation index to analyze the results before and after calibration under different atmospheric stability conditions.
[0152] The results show that the DE-SQP joint optimization framework can significantly reduce prediction error on the training set, with an RMSE reduction rate of over 96%, verifying the effectiveness of the method in model parameter correction.
[0153]
[0154] Table 7 Comparison of RMSE before and after model calibration under different atmospheric stability conditions
[0155]
[0156] 2) Test set validation and generalization ability analysis
[0157] To verify the generalization ability of the corrected model, wind turbine power data at downstream distances of 5D, 9D, and 13D that were not involved in the parameter correction process were selected to form a test set for evaluation. Figure 6 The study demonstrates the comparison between the model's predicted power and the CFD measured values under three atmospheric stability conditions, using both default and corrected parameters.
[0158] Before calibration, the model's predicted values at different locations and under different operating conditions deviated significantly from the measured values, with the maximum absolute error approaching 1100kW and the minimum absolute error being approximately 851kW. This indicates that the initial model had poor generalization prediction ability across locations and operating conditions.
[0159] After calibration, the model's prediction accuracy was significantly improved across all test locations and operating conditions, with the maximum absolute error controlled within 75kW and the minimum absolute error as low as 0.76kW. This indicates that the model calibrated using the DE-SQP joint optimization framework not only possesses high accuracy on the training set but also exhibits strong generalization ability, maintaining predictive performance in operating conditions and locations where it was not optimized.
[0160] This embodiment also includes an online self-calibration system for wake model parameters based on the DE-SQP joint framework, used to implement the above-mentioned online self-calibration method for wake model parameters based on the DE-SQP joint framework, such as... Figure 2 As shown, the system includes a wake calculation system construction module, a measured power acquisition module, a predicted power calculation module, an optimization objective function setting module, a global parameter search optimization module, a local parameter optimization module, and an application module.
[0161] in,
[0162] The system comprises the following modules: a wake calculation system construction module for building a wake calculation system including a wind turbine model, a wind speed model, and a wake model; a measured power acquisition module for acquiring measured power data of the wind turbine generator under target operating conditions; a predicted power calculation module for calculating predicted power data under corresponding operating conditions using the wind turbine model and wind speed model in the wake calculation system; an optimization objective function setting module for setting an optimization objective function based on measured and predicted power data; a global parameter search optimization module for globally searching and optimizing the multidimensional empirical parameters of the wake model using a differential evolution algorithm to obtain a globally optimal preliminary parameter vector; a local parameter optimization module for performing local fine-tuning optimization using a sequential quadratic programming algorithm with the preliminary parameter vector as the initial point to obtain the final corrected parameters; and an application module for applying the final corrected parameters to the wake model in real time.
[0163] The above-described embodiments are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Therefore, any changes made in accordance with the shape and principle of the present invention should be covered within the protection scope of the present invention.
Claims
1. A method for online self-calibration of wake model parameters based on the DE-SQP joint framework, characterized in that, include: Construct a wake calculation system that includes a wind turbine model, a wind speed model, and a wake model; The measured power data of the wind turbine generator set under the target operating condition is obtained, and the predicted power data under the corresponding operating condition is calculated using the wind turbine model and wind speed model in the wake calculation system. Based on measured power data and predicted power data, an optimization objective function is set; Based on the objective function, the differential evolution algorithm is used to perform global search optimization on the multidimensional empirical parameters of the wake model, obtain the global optimal preliminary parameter vector, avoid getting trapped in local optima and provide a good initial optimization point; Starting with the initial parameter vector, a sequential quadratic programming algorithm is used for local fine-tuning, and its fast convergence characteristic is used to obtain the final correction parameters. The final correction parameters are applied to the wake model in real time to complete the online self-calibration of the model parameters.
2. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 1, characterized in that, The wind turbine model is a simplified model of a wind turbine generator consisting of a power curve and a thrust coefficient curve. This model is fitted using field measurement data or performance curves provided by the manufacturer to establish the equivalent wind speed at the wind turbine rotor. With predicted power Thrust coefficient The mapping relationship between them is used to evaluate the power generation performance of a single wind turbine and the entire field; Among them, the equivalent wind speed at the wind turbine rotor With predicted power The mapping relationship is as follows: ； This is a power lookup table function; Equivalent wind speed at the wind turbine impeller With thrust coefficient The mapping relationship is as follows: ； This is a lookup function for the thrust coefficient.
3. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 2, characterized in that, The wind speed model uses the average wind speed from multiple sampling points as the equivalent wind speed at the wind turbine rotor. ; The sampling points are distributed in concentric circles on the hub section, and the inner and outer radii are set according to the blade length and structural characteristics. Equivalent wind speed at the wind turbine impeller The calculation formula is: in, No. Wind speed collected at each sampling point This represents the number of sampling points.
4. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 2, characterized in that, The wake model is a double Gaussian wake model, used to predict wake velocity loss. The relationship between the velocity loss at a point in the wake region and its spatial location is as follows: in, This represents the speed loss value. Downstream distance; Horizontal distance; Vertical distance; The incoming wind speed at the height of the wheel hub; The amplitude coefficient; The extreme radius of the double-peaked wake; The standard deviation of the wake width; It is a natural exponential function; Radial distance: Wheel hub height; Amplitude coefficient From thrust coefficient Hub diameter The calculation formula is determined by both the distance to the downstream side and the distance to the upstream side, as follows: in, , These are two transition parameters, expressed as follows: in, It is the error function; The standard deviation of the wake width, and its distance from the downstream. The relationship is as follows: Extreme radius of the double-peaked wake The calculation formula is as follows: , , The parameters for fitting the wake are calculated using the following formula: in, For turbulence intensity, , , , , , , , , , , , , , , These are empirical parameters; once these 15 empirical parameters are determined, the wake model can be uniquely determined.
5. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 4, characterized in that, The root mean square error between the measured power data and the predicted power data As an optimization objective: in, For the first Measured power at each location; For parameter vector based Predicted power .
6. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 5, characterized in that, The differential evolution algorithm is used to perform a global search optimization of the multidimensional empirical parameters of the wake model, including: Generate by An initial population consisting of individuals Each individual in the initial population Each corresponds to a set of parameter vectors. During the mutation phase, for the first... Target individuals in the population Generate mutation vectors using a difference strategy: in, Three distinct individuals were randomly selected from the initial population. To control the variation factor of the difference amplitude; the crossover operation will change the variation vector. With target individuals Dimensionally blended to generate test vectors For the test vector of the th Dimensional components: in, For crossover probability, For random probability, It is a randomly selected dimension index. For the mutation vector The dimensional components, For target individuals The Dimensional components; in the selection operation, by comparing the trial vectors With target individuals The fitness of the population determines which better individuals are retained for the next generation. The fitness function is the objective function. The algorithm terminates and outputs the initial parameter vector that represents the current global optimum when the maximum number of iterations is reached or the fitness change is less than a set threshold. .
7. The online self-calibration method for wake model parameters based on the DE-SQP joint framework according to claim 6, characterized in that, With the initial parameter vector Starting from point , a sequential quadratic programming algorithm is used for local fine-tuning optimization. At the th... In the next iteration, based on the current parameters Constructing QP subproblems: in, The search direction vector, The BFGS approximation of the Hessian matrix of the objective function. This represents the objective function that needs to be minimized. Represents the gradient operator. The total number of constraints. It is the transpose operator. Given the parameter constraints, solve this QP subproblem to obtain the optimal search method. Then, the step size independent variable is obtained through one-dimensional line search. Determine the first Step size of the next iteration: Update parameter vector The iterative process terminates when the first-order necessary condition of KKT is met. in, For the Lagrange multipliers corresponding to the inequality constraints, This is a locally optimal solution; Finally, the correction parameters that satisfy the physical constraints and minimize the prediction error are obtained.
8. An online self-calibration system for wake model parameters based on the DE-SQP joint framework, used to implement the online self-calibration method for wake model parameters based on the DE-SQP joint framework as described in any one of claims 1-7, characterized in that, It includes a wake calculation system construction module, a measured power acquisition module, a predicted power calculation module, an optimization objective function setting module, a global parameter search optimization module, a local parameter optimization module, and an application module; in, The wake calculation system construction module is used to construct a wake calculation system that includes a wind turbine model, a wind speed model, and a wake model. The measured power acquisition module is used to acquire measured power data of the wind turbine generator set under the target operating conditions; The predicted power calculation module uses the wind turbine model and wind speed model in the wake calculation system to calculate the predicted power data under the corresponding operating conditions. The optimization objective function setting module sets the optimization objective function based on measured power data and predicted power data; The parameter global search optimization module uses the differential evolution algorithm to perform global search optimization on the multidimensional empirical parameters of the wake model to obtain the globally optimal preliminary parameter vector. The parameter local optimization module uses the initial parameter vector as the starting point and employs a sequential quadratic programming algorithm to perform local fine optimization to obtain the final correction parameters. The application module applies the final correction parameters to the wake model in real time.