基于Adaboost的BP神经网络改进算法在短期风速预测中的应用

Application of Adaboost-Based BP Neural Network for Short-Term Wind Speed Forecast

进行较准确的风速预测对含大规模风电场的电力系统进行经济调度具有重要意义。针对目前神经网络法、时间序列法、卡尔曼滤波法等算法在短期风速预测上精度不高的缺陷,引入Adaboost算法对前馈(back propagation,BP)神经网络算法进行改进,提出了基于Adaboost的BP神经网络算法,并将该方法应用于短期风速预测。经算例分析,该算法在超前1 h和2 h的风速预测精度优于其他2种算法,且该算法在高风速段(10 m/s以上)平均绝对百分比误差低于7.5%,具有较高的工程应用价值。

It is significant for economic dispatching of power grids containing large-scale wind farms to forecast wind speed more accurately.In allusion to the defect of insufficient accuracy in current short-term wind speed forecasting by neural network,auto-regressive moving-average（ARMA） time series analysis,Kalman filtering and so on,the Adaboost algorithm was led in to improve back propagation（BP） neural network algorithm,and an Adaboost-based BP neural network method was proposed and applied to short-term wind speed forecasting.Results of analyzing calculation example showed that using the proposed Adaboost-based BP neural network the accuracy of one or two hour-ahead wind speed forecasting was superior to respective forecasting accuracy by neural network and ARMA time series analysis,and the mean absolute percentage error of wind speed forecasting by the proposed algorithm was lower than 7.5% in high wind speed period（higher than 10 m/s）.Thus the proposed method is applicable in engineering application.