【摘要】 设计了两种预测模型:利用天气变量作为输入的传统多变量预测模型;利用历史功率数据作为输入的新型单变量预测模型.采用最小冗余最大相关(min-redundancy and max-relevance,mRMR)方法分别对两种模型进行特征提取,并选用在时间序列预测方面具有优势的回声状态网络(echo state network,ESN)对未来5 min的光伏功率进行仿真预测.仿真结果表明,采用mRMR方法对历史光伏功率数据进行特征提取,确定能够使预测模型达到最优效果的特征子集,并将其作为单变量预测模型的输入,可以得到更准确的预测效果.所构建的新型单变量预测模型能够为光伏电站提供新的光伏预测思路.更多还原

【Abstract】 Two kinds of prediction model are designed,ie,the traditional multivariable forecasting model using weather variables as input and the new univariate forecasting model using historical power data as input.The mRMR(minimum-redundancy and maximum-relevance)method is used to extract the features of the two models respectively,and ESN(echo state network),which has advantages in time series prediction,is used to simulate and predict the photovoltaic(PV) power in the next 5 minutes.The simulation results show that the mRMR method is used to extract the features of the historical PV power data,then,the feature subset that can make the prediction model reach the optimal effect is determined.The feature subset is used as the input of the univariate prediction model to obtain more accurate prediction effect.The new univariate prediction model can provide a new PV prediction idea for PV power station.更多还原