基于复杂过程简化模型的DHP学习控制  

Learning Control of DHP Method Based on Complex Process Simplified Model

作　　者：陈宗海[1] 文锋[1] 

机构地区：[1]中国科学技术大学自动化系,合肥230027

出　　处：《控制与决策》2006年第10期1087-1091,共5页Control and Decision

基　　金：国家自然科学基金项目(60575033)

摘　　要：提出一种基于简化模型的DHP(Dual Heuristic Programming)方法的学习控制,避免了标准DHP方法需要被控对象的精确模型来求得对于状态和控制动作的Jacobian矩阵,而是利用简化过程对象模型获得近似Jacob ian矩阵,实现学习控制的需要.生化反应器定值控制的仿真结果表明,该方法加快了学习过程,并对更大范围的参数变化具有鲁棒性.The standard DHP method needs accurate plant model to calculate the Jacobian matrix of state and control action, which is difficult to meet. A learning control strategy based on DHP（Dual Heuristic Programming） method of simplified model is proposed, which applies approximate Jacobian matrix to DHP training and thus relaxes this limitation. Simulation results of contrapose Bioreactor show that the proposed method can accelerate learning process and is robust to larger parameter changes.

关 键 词：强化学习 DHP方法 生化反应器 简化模型 

分 类 号：TP13[自动化与计算机技术—控制理论与控制工程]