Echo State Network with Information Inheritance Properties

Chaotic systems are highly sensitive to initial conditions, and the extreme unpredictability of their dynamic behavior poses significant challenges for precise prediction. Original Echo State Network (OESN) has limitations in feature extraction and is insufficient for hierarchical modeling when applied to chaotic system prediction. To address this issue, this paper proposes an Echo State Network with Information Inheritance properties (In-ESN), constructing a lightweight hierarchical architecture through an innovative cross layer dynamic coupling mechanism. The model employs a double reservoir structure, where the collaborative work of the information layer and the decision layer enables the progressive extraction and fusion of multilevel features. Experimental results show that In-ESN demonstrates superior performance in the prediction tasks for multiple classical chaotic systems, verifying its significant advantages in feature extraction and long-term prediction.