偏鲁棒M回归回声状态网络

Partial Robust M-regression Echo State Network

为了提高回声状态网络的非线性映射能力和网络的学习性能,提出了偏鲁棒M回归回声状态网络。首先,将储备池激活函数的输出矩阵作为PRM算法的输入样本数据,输出向量作为PRM算法的输出样本数据;其次,对输入输出样本进行加权处理,建立它们之间的回归模型来获取PRM算法的回归系数(即ESN网络的输出权值);最后,通过仿真实验验证,与回声状态网络相比,该算法不仅是有效的、可行的,而且具有较高的测试精度和良好的泛化能力。

In order to improve the nonlinear mapping capability and learning performance of echo state network,a new learning algorithm called partial robust M-regression echo state network is proposed. Firstly,the output matrix of reservoir activation function is the input sample data for the PRM algorithm,the output vector is the output sample data of PRM algorithm. Secondly,the weight of the sample data is calculated. In order to obtain the regression coefficients of PRM algorithm(that is,the output weights of ESN network),a regression model between them is established.Finally,the effectiveness and feasibility of the algorithm is verified using simulation experiment. Compared with the ESN,the simulation results show that the algorithm has better prediction precision and good generalization performance.