摘要
针对风电场中邻近多台风电机组集中出现缺损测量风速的工况,提出基于粒子群优化广义回归神经网络的风电机组缺损测量风速集成填充方法。以"成员等同性"原则引入动态时间规整算法、空间邻点法和Pearson相关系数法,分别搜寻与缺损测量风速风电机组风速演化最为相似的若干台风电机组及对应的测量风速时序,建立基于广义回归神经网络的填充子模型,采用粒子群算法对广义回归神经网络的模型参数和训练集的构成进行全局优化,之后选取较好的子模型构造自适应的熵权集成填充模型。实验结果表明:依据相似性风速序列进行缺损风速的填充能有效提高填充精度;粒子群算法优化广义回归神经网络,不仅提高了子模型的填充效果,更使得模型参数的调节有据可依,能适应不同风电场风速数据的特点;基于熵权的集成填充策略理论依据充分,集成填充的精度和稳定性优于单个子模型。
For the working condition of adjacent wind turbines being missing measurement data of wind speed in wind farm, an ensemble interpolation method of missing measurement data of wind speed based on Particle Swarm Optimized Generalized Regression Neural Network （PSO-GRNN）was proposed. Firstly, on the basis of equal-likelihood of the sub- models, Dynamic Time Warping （DTW）method, Spatial Nearest Neighbor （SNN）method and Pearson Correlation Coefficients （PCC）method were introduced to measure the similarity of wind speed timing series between the turbine missing wind speed and other turbines in the wind farm, and then 3 candidate models were constructed. Secondly, the PSO algorithm was used to optimize the parameters and the composition of the training set of the GRNN. Lastly, an ensemble interpolation model based on entropy weight was constructed using the superior candidate models. The experiment results show that the accuracy of the interpolation model can be greatly improved by using the similar wind speed data. The PSO algorithm not only further improves the interpolation effect, but makes the adjustment of the model parameters more praetical, and more adaptable to the characteristics of wind speed data in different wind farms. The ensemble model based on entropy weight has a sound theoretical basis, and the accuracy and stability of the ensemble model are superior to the single sub-model.
出处
《太阳能学报》
EI
CAS
CSCD
北大核心
2016年第8期2104-2110,共7页
Acta Energiae Solaris Sinica
基金
国家自然科学基金(61103142
41105059
41575070)
科技部公益性行业(气象)科研专项(GYHY201306002)
江苏高校优势学科建设工程
关键词
缺损测量风速填充
风电机组
粒子群
广义回归神经网络
熵权集成模型
interpolation missing wind speed data
wind turbine
particle swarm optimizing
generalized regression neural network
ensemble model based on entropy weight
