A novel reservoir design strategy for echo state network

In conventional echo stat network (ESN), the reservoir are randomly generated, then the spectral radius of the reservoir is scaled to lower than 1. In this method, only the necessary condition for echo state property (ESP) of ESN is satisfied while the sufficient condition is ignored, thus the ESN stability may not be ensured. In this paper, with the predefined singular values (smaller than 1), the reservoir is constructed by using singular value decomposition (SVD) method, and the sufficient condition for ESP is guaranteed. The simulation results show that the learning ability and prediction accuracy of the proposed ESN are improved simultaneously.