Parameter optimization of phase space reconstruction for short-term load time series

The delay time τ, embedded space dimension m and reference points have important influence on short-term load forecasting by the local linearization method of phase space reconstruction. The delay time window Γ is determined by the attractor appearance and autocorrelation analysis of power short-term load time series. Based on the fundamental relation of m, τ and Γ, several sets of optimal combinations of m and τ are advanced for power loads, and the combinations of m and τ are proved to be invariable with load's linear transformation. An effective method composed of rough search and fine search is presented to choose reference points. The rough search is used to search some neighboring points at first, and some false neighboring points are kicked off through fine search in terms of the time evolution relativity. The method is then applied to practical load forecasting, the result shows that the forecasted load data are more accurate.