Time series prediction based on ring echo state network with sine distribution input
LUN Shuxian
ZHANG Ao
Abstract：Echo state network(ESN)represent a relatively novel class of recurrent neural networks that have demonstrated significant potential in chaotic time series prediction.However,the random generation of ESN'connectivity and weight structures may lead to suboptimal reservoir performance during task execution.Additionally,the input weight matrix may exhibit substantial fluctuations during random initialization.To address these limitations,we have developed a novel architecture termed sine-circle echo state network(SCESN),which features a deterministic reservoir structure and an input weight matrix with autocorrelation properties.The SCESN employs a sinusoidal distribution for the input weight matrix to ensure both sparse connectivity of the reservoir and autocorrelation of input weights.Furthermore,the reservoir structure is designed based on the Newman-Watts small-world network topology.Comparative experiments across multiple datasets demonstrate that the SCESN outperforms traditional ESN in prediction accuracy while maintaining lower network complexity.
Keywords：echo state networkreservoirchaotic time series predictionnetwork topology design
Publication Date：2025-09-15
Online Publishing Date：2025-12-22(First online date of this platform, not the publication date of the document)
Pages：10( 231-240 )