Enhanced rotating neuron reservoir computing with time-delay feedback for chaotic time series prediction

Chaotic time series prediction has attracted considerable attention due to its wide-ranging applications in atmospheric turbulence modeling, early warning systems for financial market fluctuations, and secure chaotic communication. Rotating neuron-based architectures are often employed in this domain because of their structural simplicity and low computational complexity. However, a major limitation of existing rotating neuron models lies in the accumulation of errors during recursive prediction, which restricts the attainable prediction horizon. To overcome this limitation, this paper introduces an enhanced rotating neuron reservoir computing architecture that integrates time-delay feedback dynamic neurons and an adaptive error feedback mechanism to suppress error accumulation. The effects of key hyperparameters on prediction performance are also systematically investigated. Simulation results demonstrate that the proposed system achieves continuous chaotic time series prediction over 1.5 ns with a normalized root mean square error below 0.1. By combining the dynamic properties of time-delay feedback neurons with the structural simplicity of rotating neurons and incorporating adaptive error correction, the proposed architecture offers an efficient and scalable solution for chaotic time series forecasting. This method shows good potential for continuous prediction tasks in optical systems and other chaotic dynamical environments.