Time Series Prediction Using Time-Series Decomposition and Multi-reservoirs Echo State Network

Abstract

AbstractEcho State Network (ESN) is effective to do time series analysis, which has a dynamic reservoir, includes: input units, internal units and output units. However, due to the randomness and non-stationarity properties of most time series, it is difficult for single reservoir to better handle them, because different scales in the time series should be dealt with in a unified structure. In order to solve this, the time-series decomposition (TSD) is employed to decompose the time series into different sub-sequences, which can be handled by different reservoirs. Now, there are many TSD methods, such as: empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD), complementary ensemble empirical mode decomposition (CEEMD), local mean decomposition (LMD), variational mode decomposition (VMD), etc. Different TSD methods are used to decompose the time series and then the multi-reservoirs ESN are constructed. Finally, experimental results using several time series are presented and compared.