基于贝叶斯回归的多核回声状态网络研究

Research on multi-reservoir ESN based on Bayesian regression

在利用单储备池模型对多变量预测研究时,多个变量只能通过单个储备池进行特征映射,无法分别刻画各个变量的动力学特性.针对以上问题,提出一种多储备池回声状态网络.混沌系统中各个变量分别通过各个储备池扩展成高维的特征向量,采用Bayesian线性回归的方法,对多核回声状态网络输出权值进行训练,形成一种新的预报器,即多核贝叶斯状态回声机(MrBESN).实际数据的仿真结果验证了所提方法的有效性.

When multi-variable time series are predicted with single reservoir model,the dimensions of phase-space reconstruction can be only selected a single value,which can not portray respectively the dynamic feature of complex system. Therefore,multi-reservoir echo state network is presented,which can expand the simple input into high-dimesional feature vector.Multi-reservoir echo state network is presented which is based on Bayesian regression.Realistic simulation results show the effectiveness of the predictor proposed.