教与学优化储备池计算的混沌时间序列预测  

Chaotic Time Series Prediction Using Reservoir Computing Via Teaching-Learning-Based Optimization

作　　者：李志刚 高媛媛 孙晓川 LI Zhi-gang;GAO Yuan-yuan;SUN Xiao-chuan(College of Artificial Intelligence,North China University of Science and Technology,Tangshan 063200,China)

机构地区：[1]华北理工大学,河北唐山063200

出　　处：《计算机仿真》2022年第5期455-460,共6页Computer Simulation

摘　　要：针对回声状态网络初始化参数的随机性影响网络预测精度的问题,提出一种利用教与学优化算法对网络的输入权值、储备池内部权值和输出权值进行局部优化的预测模型,保留了储备池网络权值参数的部分随机性,从而具有良好的泛化能力。为验证本模型的有效性,针对Mackey-Glass和Lorenz混沌时间序列的预测任务进行仿真。由仿真结果可知,对比相关粒子群优化方法,教与学优化的回声状态网络模型在预测性能、收敛速度和稳定性方面有显著的优势。Aiming at the problem that the parameter randomness of echo state network(ESN) affects the prediction accuracy,the Teaching-learning-based optimization algorithm is applied to optimize the input weight,the reservoir weight and the output weight of the reservoir,which retains part of the randomness of this model and has good generalization ability.To verify the validity of this model,the prediction task of Mackey-glass and Lorenz chaotic time series prediction tasks was simulated.Simulation results show that our proposal has more significant advantages in prediction,convergence and stability performance compared with particle swarm optimization methods.

关 键 词：储备池计算 教与学优化 混沌时间序列预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]