基于可变模式分解和NWCSO优化极限学习机的短期风速预测

Short-Term Wind Speed Forecasting Based on Variational Mode Decomposition and Extreme Learning Machine Optimized by NW Small World Crisscross Optimization

针对风速时间序列的非线性特征导致其难以准确预测的问题,提出一种基于可变模式分解(variational mode decomposition,VM D)和动态NW小世界纵横交叉算法(dynamic NW small w orld crisscross optimization,NWCSO)优化极限学习机的短期风速组合预测模型。采用一种新型的可变模式分解技术,将原始风速时间序列分解为一系列不同带宽的模式分量以降低其非线性,然后对全部分量分别建立极限学习机模型进行预测,并采用小世界纵横交叉算法对极限学习机的输入权值和隐含层偏置进行优化,以获得最佳的预测效果。实验结果表明,基于VMD的组合预测模型较采用其他常规分解方式时预测精度明显提高。

Aiming at the problem that wind speed time series are always nonlinear so that it is difficult to predict accurately, this paper proposes a combined model based on variational mode decomposition （VMD） and extreme learning machine （ELM） optimized by dynamic NW small world crisscross optimization （NWCSO） for short-term wind speed prediction. We used a novel VMD technique to decompose the original wind speed time series into a series of mode components with different bandwidth to reduce its non-linearity. Then, we established the extreme learning machine for predicting all components, and adopted the NWCSO to optimize the input weights and the bias of hidden nodes of ELM in order to obtain the best prediction effect. The results have demonstrated that the proposed model based on VMD has higher prediction accuracy than other conventional decomposition methods.