基于数据驱动方法的风电机组功率优化

摘要

To improve the wind energy production and profit,it is imperative to optimize the power output of wind turbine generator (WTG).This paper extracts the analytical relation between WTG power,wind profile and control variables from historical operation data with a designed feed-forward neural network.Based on such a relation,point-to-point and cluster optimization strategies are developed and used for WTG power optimization,which optimize WTG control variables for maximum WTG power under the wind profile measured.The K-means clustering algorithm is used in the latter strategy to reduce optimization complexity,thus facilitating real-time WTG power optimization.Three new indices of mean power gain (MPG),rate of power gain (RPG) and probability of power gain (PPG) are proposed to quantify the power gains by the optimization strategies proposed.Extensive comparisons are conducted between two proposed strategies and recorded operation data using H56-850 WTG.Results show that both strategies can optimize WTG power output.In addition,cluster strategy with five clustering centers could considerably reduce optimization complexity while achieving similar effectiveness as that of the point-to-point strategy.%风电机组功率与风电产能息息相关，为提高风电产能及其收益，有必要对风电机组功率进行优化。应用前馈神经网络，从历史运行数据中挖掘风电机组功率与风速和控制量间的函数关系，进而提出逐点优化策略和聚类优化策略，用于实现风电机组功率优化，即在已知测量风速时，优化确定风电机组控制量，实现风电机组功率最大化。后者优化策略在前者优化策略基础上，应用K 均值聚类方法聚类风速，从而降低优化计算复杂度，利于风电机组功率的实时优化。定义平均功率增益、功率增益百分比和功率增益概率三种指标用于测度功率优化效果。将两种优化策略应用至H56-850型风电机组，将优化后的风电机组功率与历史运行记录进行对比，结果表明，两种优化策略均可有效提高风电机组功率输出。此外，聚类中心数为5的聚类优化策略，能以较低的优化计算复杂度，达到与逐点优化策略相近的优化效果。