CN-121978943-A - Model prediction yaw control method and system based on distributed robust optimization

CN121978943ACN 121978943 ACN121978943 ACN 121978943ACN-121978943-A

Abstract

The invention relates to the technical field of wind generating set control, and discloses a model prediction yaw control method based on distributed robust optimization. S1, acquiring historical wind direction time sequence data of a wind turbine generator, preprocessing the historical wind direction time sequence data by adopting a continuous variation modal decomposition method to reconstruct middle-low frequency wind direction data, and training a converter prediction network based on the reconstructed middle-low frequency wind direction data to obtain a high-precision wind direction prediction model. According to the method, the original wind direction sequence is subjected to self-adaptive denoising and reconstruction by adopting continuous variation modal decomposition, so that the problem of modal aliasing in the traditional method is solved, and the regularity and stability of time sequence characteristics are enhanced. According to the invention, the distributed robust optimization framework is introduced into the model prediction yaw control, and the optimization decision can be carried out on the premise of not depending on the accurate error probability distribution, so that the smaller yaw error can be maintained under the high uncertainty scene with larger error in wind direction prediction, and the robustness of the control system is improved.

Inventors

QI XIAO
WANG WENHAO
DENG HUI
ZHANG ZHIGANG
NIU HAIMING
LI JIALIN

Assignees

暨南大学
国能智深控制技术有限公司

Dates

Publication Date: 20260505
Application Date: 20260203

Claims (10)

1. A model predictive yaw control method based on distributed robust optimization, the method comprising: s1, acquiring historical wind direction time sequence data of a wind turbine generator, preprocessing the historical wind direction time sequence data by adopting a continuous variation modal decomposition method to reconstruct middle-low frequency wind direction data, and training a converter prediction network based on the reconstructed middle-low frequency wind direction data to obtain a high-precision wind direction prediction model; S2, constructing a model prediction yaw control frame based on a multi-step wind direction prediction result output by the high-precision wind direction prediction model, wherein the yaw control frame aims at minimizing a multi-objective optimization function comprising power loss and yaw loss, and models wind direction prediction uncertainty as a Wasserstein fuzzy set based on historical prediction error experience distribution; s3, decomposing a distributed robust optimization problem based on the Wasserstein fuzzy set in the model prediction yaw control framework into a plurality of sub-problems which can be solved in parallel by adopting a decomposition solving method to obtain an optimal yaw control instruction of the current control period; and S4, applying the optimal yaw control instruction to a yaw actuating mechanism of the wind turbine generator, and returning to the execution step S2 in the next control period to realize rolling optimization and feedback control.
2. The method according to claim 1, wherein the step S1 specifically comprises: dividing the historical wind direction time sequence data into a training set, a verification set and a test set; Performing self-adaptive modal decomposition and selective reconstruction on wind direction data in the training set and the verification set by adopting a continuous variation modal decomposition method to obtain reconstructed medium-low frequency wind direction data; Training and verifying the constructed converter prediction network by taking the reconstructed medium-low frequency wind direction data as a training sample to obtain the high-precision wind direction prediction model; Predicting a trained prediction model by using wind direction data of the verification set, and collecting prediction errors to construct the historical prediction error experience distribution used for representing wind direction prediction uncertainty.
3. The method according to claim 1, wherein in the step S2, the multi-objective optimization function J is: ; Wherein, the To normalize the power loss term, and yaw error Cosine square correlation of (a); a yaw motion time loss term for normalization; A normalized yaw motion number loss term; 、、 is the corresponding weight coefficient.
4. The method according to claim 1, wherein in the step S2, constructing the wasperstein fuzzy set specifically includes: determining a reference distribution for each predicted time step based on the historical prediction error empirical distribution ; With the reference distribution Centrally, constructing fuzzy sets meeting Wasserstein distance constraint The fuzzy set Containing all possible candidate true distributions The mathematical expression is as follows: ; ; Wherein, the For the wasperstein distance, For the total wasperstein radius, m is the prediction step, And The candidate true distribution and the reference distribution of time step t respectively, Is a joint probability distribution.
5. The method according to claim 4, wherein in the step S2, the distributed robust optimization problem is expressed as a min-max optimization problem: ; Wherein, the Is a sequence of yaw motions, Is a yaw motion enabled domain, Is a true distribution of the candidates, Based on candidate true distribution The formula is expressed as finding the worst distribution that maximizes the total expected loss among all potential true distributions From the feasible yaw motion sequence And screening the optimal yaw motion sequence that minimizes the expected total loss for the worst distribution.
6. The method according to claim 5, wherein in step S2, the construction of the yaw action feasibility field D satisfies the following constraints: yaw rate Taken from a finite set , wherein, Represents a reverse yaw of-0.5, Indicating a forward yaw of 0.5 °,0 indicating no yaw; And in the prediction time domain, the switching times of the control quantity symbols in the yaw action sequence do not exceed a preset threshold value.
7. The method according to claim 1, wherein the decomposition solving method in step S3 specifically includes: The total Wasserstein radius Dynamically distributing to each predicted time step to obtain the radius of each time step ; Converting the min-max optimization sub-problem of each time step into a quadratic programming problem based on a dual principle; by enumerating all possible yaw action sequences Solving the quadratic programming problem corresponding to each time step in parallel, and calculating the expected total loss of each sequence under worst distribution; And selecting a yaw motion sequence which minimizes the expected total loss as an optimal sequence, and taking a first control quantity of the yaw motion sequence as the optimal yaw control command.
8. The method of claim 7, wherein the total Wasserstein radius is Dynamically distributing to each prediction time step, specifically: according to the amplitude and time sequence of the short-term prediction error, each time step is provided Assigning weights ; According to the formula Calculating dynamic radius of each time step Wherein time steps with a recent or large prediction error are assigned larger radii.
9. The method according to claim 1, wherein the step S4 specifically includes: transmitting the optimal yaw control instruction to an electric actuator of a yaw system to drive a nacelle of the wind turbine to deflect; When the next control period comes, the wind direction and the cabin position are re-measured, and the system state is updated; And (3) repeatedly executing the steps S2 and S3 based on the updated state, generating a new optimal yaw control instruction, and realizing closed-loop rolling optimal control.
10. A model predictive yaw control system based on distributed robust optimization for implementing the model predictive yaw control method based on distributed robust optimization of any one of claims 1-9, the system comprising: The data acquisition and prediction module is used for acquiring wind direction data, executing continuous variation modal decomposition and transform network prediction, and outputting a multi-step wind direction prediction result and error experience distribution thereof; the optimization model construction module is used for constructing a multi-objective distribution robust optimization model containing Wasserstein fuzzy sets based on the prediction result; the online solving and deciding module is used for solving the optimizing model in real time by adopting a decomposition method and outputting an optimal yaw control instruction; The instruction execution and feedback module is used for sending the control instruction to the yaw executing mechanism, collecting a new system state and feeding back the new system state to the optimization model construction module; the online solving and deciding module further comprises: The radius dynamic allocation unit is used for adaptively adjusting the Wasserstein fuzzy set radius of each prediction time step according to the short-term prediction error; and the parallel solving unit is used for decomposing the high-dimensional optimization problem into a plurality of quadratic programming sub-problems to perform parallel calculation.

Description

Model prediction yaw control method and system based on distributed robust optimization Technical Field The invention relates to the technical field of wind generating set control, in particular to a model prediction yaw control method and system based on distributed robust optimization. Background Under the promotion of global energy transformation, wind power generation is an important component of clean renewable energy, and stable and efficient operation of the wind power generation has important significance for power grid safety and energy supply. The output power and the running stability of the wind turbine generator are highly dependent on the control performance of the yaw system. The yaw system dynamically tracks the change of wind direction to ensure that the wind wheel faces the incoming wind direction, so that the maximum capture of wind energy is realized in the whole wind speed range. Therefore, the accuracy and the adaptability of the yaw control strategy are directly related to the power generation efficiency, the equipment loss and the service life of the wind turbine generator. Currently, in yaw control of a wind turbine generator, a fixed threshold control strategy is mostly adopted in the traditional method, namely whether yaw is started or not is determined according to a preset yaw error threshold value and delay time. The method is simple and reliable, but has the advantages that parameters are fixed, the wind speed interval is divided coarsely, the method is difficult to adapt to complex and changeable wind conditions, yaw motion is frequent, wind accuracy is insufficient or response is delayed, the power generation efficiency is affected, and mechanical abrasion of a yaw system is increased. In order to improve yaw control performance, various intelligent optimization and predictive control methods are introduced into the field. For example: According to the parameter optimization method based on the improved genetic algorithm disclosed in the patent document with the publication number of CN109340046A, through analyzing historical operation data, yaw error threshold values and delay time are optimized according to wind speed segments, so that the action frequency can be effectively reduced, but the optimization process depends on historical data distribution, and the method belongs to offline or periodic optimization, and is difficult to respond to second-level rapid change of wind direction in real time. The method based on model predictive control disclosed in the patent document with publication number CN114263565A utilizes a future wind direction prediction sequence, aims at maximizing the generated energy and optimizes a yaw action sequence in an online rolling way, so that the wind accuracy and the generated electricity income are remarkably improved, but the performance of the method is highly dependent on the accuracy of wind direction prediction, and the control effect is easy to deteriorate in a scene with high prediction uncertainty. The data-driven strategy self-adaptive adjustment method disclosed in the patent document with the publication number of CN116696669A dynamically adjusts control parameters by analyzing the distribution characteristics of the running time of a unit, improves the adaptability of the strategy, but the adjustment mechanism is mostly based on statistical analysis, and lacks modeling and robust processing of a system for inherent randomness and structural errors of wind direction prediction. In summary, when the existing yaw control method is used for coping with random fluctuation of wind direction and uncertainty of prediction, especially for offshore complex wind conditions, the existing yaw control method still has the following common defects that firstly, the performance is reduced when prediction is inaccurate due to the fact that most methods lack of explicit modeling and robust decision mechanisms for wind direction prediction errors, secondly, although future information is utilized by the existing prediction control method, multi-objective optimization of power generation benefits and yaw losses (such as motor abrasion and brake pad loss) cannot be effectively balanced, and thirdly, the method for deeply fusing high-precision wind direction prediction, prediction uncertainty quantification and distributed robust decision is not yet achieved, so that yaw control with high precision, strong robustness and instantaneity is achieved. Therefore, aiming at the problems that the existing yaw control method is limited in prediction precision, lacks a wind direction prediction uncertainty processing mechanism and is difficult to adapt to the dynamic change of complex wind conditions, a robust prediction control method capable of integrating high-precision wind direction prediction, explicit modeling prediction uncertainty and maintaining performance in the worst case is needed to be researched, so that the wind energy capturing efficiency and