基于ESN的非线性随机系统对偶自适应控制

Dual adaptive control of nonlinear stochastic systems based on ESN

设计了一种用于一类单入单出(SISO)离散随机仿射非线性系统的自适应对偶控制器.回声状态网络(ESN)是一种带有动态池的回归神经网络,通过使用卡尔曼滤波在线调节ESN的参数,估计未知非线性系统的模型,然后基于一种显式次优的代价函数来设计其对偶自适应控制律.最后通过仿真及蒙特卡罗分析验证了所提出的对偶控制律的有效性.

The dual adaptive control problem was addressed for a class of single‐in‐single‐out （SISO） stochastic ,affine nonlinear ,discrete systems .The nonlinear functions of system model were assumed to be unknown and approximated by the echo state network （ESN） ,which were recurrent neural net‐works with dynamic reservoirs .The parameters of ESN were online adjusted by using the convention‐al Kalman filtering technique .The dual adaptive control law was designed considering an explicit‐type ,suboptimal cost function based on the innovations .The performance of proposed control law was verified by simulations and Monte Carlo analysis .