Direct multi-step prediction of wind speed based on chaos analysis and DRNN
Liu Xingjie, Yanqing Zhang, Zengqiang Mi, Fan Xiaowei, WU Jun-hua · 2009
The direct multi-step prediction employs measurement data but not the results of single-step prediction. So it should have a better prediction effect for short-term wind speed. Aiming to the chaotic nature of wind speed data, a novel direct multi-step prediction approach for wind speed has been presented in this paper. This approach was based on chaos analysis and dynamic recurrent neural network(DRNN). According to the phase space reconstruction theory, the phase space of wind speed data was first reconstructed. As a result, the attractor reflecting the inner rules of wind speed data was obtained. Then a DRNN model was established on the basis of the attractor. After training the network, the built model was used to directly predict the wind speed ahead of some steps. The detailed procedures were introduced in this paper. The results of a wind farm's simulation show that the proposed approach greatly improves the multi-step prediction accuracy.