Research on hierarchical modular ESN and its application

Aim to solve the problem of structure design about echo state network, a new type of ESN with hierarchical modular structure (HMESN) is proposed in this paper. The hierarchical connections are introduced into reservoir of HMESN, and each level of neurons are modular structure. The structure of HMESN is closer to the hierarchical modular topology characteristics of brain networks, which effectively enhances the dynamics of internal neurons. Finally, the HMESN is used for the Mackey-Glass time series prediction and the sewage treatment modeling. Experimental results and the performance comparison demonstrate that the prediction accuracy of the proposed HMESN is better than those of ESN.