摘要
根据人工神经网络理论提出了一种处理最优控制问题的新思想，即由每一采样时刻系统的输入和响应，快速训练产生一模型网络，并由ＬＱＲ理论直接构造最优反馈策略．和以往研究不同，这里不需要任何离线学习，因而避免了收集离线学习所需数据的困难，可以实现实时控制。
A novel method for nonlinear discrete time optimal control problems is proposed based upon the neural network theory. The nonlinear dynamic system is approximated by a linear network model which is rapidly updated by using the system inputs and outputs during each sampling interval. The Linear Quadratic Regulator(LQR) theory is directly used to deduce the optimal control. Because no off line training is required, this new method can be applied to real time control situations. An illustrative numerical example shows that it is rather feasible and effective. This control scheme is a little more intelligent than any previous ones.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
1996年第1期34-38,共5页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金资助项目
关键词
神经网络
离散时间
最优控制
非线性系统
artificial neural networks discrete time optimal control nonlinear systems
