基于混沌分析和神经网络的风速直接多步预测

DIRECT MULTI-STEP FORECASTING OF SHORT-TERM WIND SPEED BASED ON CHAOS ANALYSIS AND NEURAL NETWORK

针对风速序列的混沌特性,提出了一种将混沌分析和神经网络相结合的短期风速直接多步预测新方法,以提高其预测精度。首先,对风速序列进行混沌特性分析和相空间重构;然后,根据重构相空间的特征参数,结合预测需求,确定Elman网络结构;最后,利用空间欧式距离选取的样本对Elman网络进行训练,建立风速直接多步预测模型。以华北地区某风电场实测风速为例进行仿真测试,结果表明与单步迭代法和直接神经网络法相比,该文方法在进行风速直接多步预测时具有更好的整体误差指标。

To improve the accuracy of direct multi-step forecasting for wind speed, a novel forecasting approach was presented aiming at the chaotic nature of wind speed data, which combined the chaos analysis and the neural network. Firstly, the phase space of wind speed data was reconstructed according to the phase space reconstruction theory. As a resuh, the attractor reflecting the inner rules of wind speed data was obtained. Then the framework of Elman network was determined on the basis of the attractor and the forecasting requirements. After training the network with testing samples chosen from the Euclidean distance, the Elman network model used to perform the direct multi-step forecasting was finally built. The detailed procedures were introduced in this paper. The simulation results for a real wind farm show that the proposed approach has a better global forecasting performance comparing to the iterative multi-step forecasting methods and the Elman network.