基于递归神经网络的移动域控制方法

Receding-horizon Control Method with Recurrent Neural Networks

构造一种线性差分式Hop fie ld网络(LDHNN),其稳定状态可使能量函数达到唯一极小值.利用该网络稳定性与其能量函数收敛特性的关系,提出了基于LDHNN的移动域控制方法.LDHNN的理论设计表明,网络的稳态输出即为移动域LQ控制问题的解.当系统满足一定条件时,基于LDHNN的移动域LQ控制能保证闭环最优控制系统的渐近稳定性.数字仿真取得了与理论分析一致的实验结果.

A linear difference Hopfield neural network （LDHNN） is built, and its energy function can reach the only minimum while LDHNN is stable. With the use of the relation between the stability and energy function convergence of the Hopfield neural network, an LDHNN-based receding-horizon （RH） control method is proposed. The theoretical design of LDHNN shows that the stable outputs of LDHNN are the solution of the RH LQ control problem. The LDHNN-based RH control can also guarantee the asymptotical stability of closed-loop optimal control systems if the controlled systems satisfy certain conditions. The numerical simulation results show the correction of theoretical analysis.