基于小波包分解和改进差分算法的神经网络短期风速预测方法

Short-term Wind Speed Forecast Method Based on WPD-IDE-NN

风速预测在风电场的智能管理和安全并网中起着至关重要的作用,针对风速预测固有的波动性、间歇性和非线性等特点,以及常规BP神经网络和差分算法神经网络均存在容易陷入局部最优导致收敛过早、泛化能力不足等缺陷,提出一种综合WPD和IDE算法的短期风速预测神经网络方法。该方法首先利用WPD将风速的时间序列分解成多种不同频率的子序列,然后采用IDE算法优化后的神经网络对小波包分解后的每个不同频率的子序列进行单步预测,最后将预测后的各个子序列进行叠加,得出最终预测结果。为验证所提方法的有效性,将其分别与采用混合小波分解的BP神经网络风速预测方法和混合小波分解的差分算法风速预测神经网络方法进行对比,对某地区的实际风速数据进行实验仿真,结果表明,所提方法的预测精度明显优于其他算法。

Wind speed forecasting is of great importance for intelligent wind farm management power system integration safety. Wind speed forecast is characterized by inherent volatility,intermittence and nonlinearity; in addition,the conventional back propagation （BP） neural network and differential evolution（DE） neural net-work tend to fall into local optimal which leads to premature convergence and weak generalization ability. The paperpresentsashort_termwindspeedforecastmethodbasedonneuralnetwork,which combines wavelet packetdecomposition （WPD） withimproveddifferentialevolution （IDE）.Firstly, WPDisemployedtodecom_ pose the wind speed time series into sub-series with different frequencies. Secondly,the IDE optimized neural is used for single-step prediction of the sub-series. The eventual predicted results are obtained through sub-se-ries aggregation. To verify the proposed method,it is respectively compared with the wind speed forecasting method based on neural networks that adopts hybrid WPD and wind speed forecasting method based on DE that adopts hybrid WPD. Experimental simulation is conducted on actual wind speed data in a specific area, which demonstrates that forecast accuracy of the proposed method is comparatively higher.