Nonlinear stochastic coupling-enhanced reservoir computing for state prediction of chaotic systems

Reservoir computing is widely used for system modeling and prediction due to its high training efficiency and low computational cost. However, traditional approaches typically rely on a single deterministic reservoir architecture and neglect nonlinear interactions among reservoir nodes, which limits the richness of reservoir dynamics. To address these issues, this paper introduces nonlinear interaction structures within the reservoir to enhance the expressive capability of its internal dynamics. Meanwhile, an inter-reservoir stochastic coupling mechanism is designed to promote information exchange and state representation across reservoirs. Based on these designs, we propose a nonlinear stochastic coupling-enhanced reservoir computing framework (NSCRC). Numerical experiments on the Lorenz-63 and Rössler benchmark systems show that NSCRC achieves substantial improvements over traditional approaches in both short-term predictive accuracy and long-term dynamical consistency. The proposed framework provides a general and efficient approach for state prediction in chaotic systems.