具有双储层结构的动态误差补偿回声状态网络

针对传统回声状态网络难以有效应对高阶非线性复杂模型问题,本文在理论分析的基础上提出了一种双储层结构的误差补偿回声状态网络,并设计了该网络的学习算法.该网络由计算层和补偿层构成,计算层主要承担拟合任务,补偿层则作为状态跟随器,实时补偿由于计算层对期望方差估计不足而导致的幅值偏差.对多阶振荡器和真实高阶非线性数据集的实验结果表明,本文所提网络结构较常规网络具有更高的稳定性和泛化性能,尤其对高阶非线性复杂模型的预测精度大幅度提升.

A new echo state network with a double reservoir compensates for dynamic error

The traditional echo state network is challenging to deal with the high-order nonlinear complex model effectively. We proposed an error trace reservoir computing and designed the optimal network algorithm. This new reservoir computing structure consists of a computing layer and a compensation layer. The computing layer mainly undertakes the fitting task, and the compensation layer acts as an error trace function. Because the computing layer always has an insufficient variance estimation, it will lead to unstable neural network prediction. Thus, we proposed the compensation layer to trace neural network error in real-time. The numerical experiments on modeling the multiple superimposed oscillators and nonlinear data sets demonstrate that error trace reservoir structure has higher stability and generalization performance than the conventional network, especially in the high order nonlinear complex models.