小脑模型关节控制器网络在传送带给料生产加工站学习优化控制中的应用

Application of cerebellar model articulation controller network to learning optimization control in conveyor-serviced production station

研究单站点传送带给料生产加工站(conveyor-serviced production station,CSPS)系统的前视(look-ahead)距离最优控制问题,以提高系统的工作效率.论文运用半Markov决策过程对CSPS优化控制问题进行建模.考虑传统Q学习难以直接处理CSPS系统前视距离为连续变量的优化控制问题,将小脑模型关节控制器网络的Q值函数逼近与在线学习技术相结合,给出了在线Q学习及模型无关的在线策略迭代算法.仿真结果表明,文中算法提高了学习速度和优化精度.

This paper is concerned with the optimization of the look-ahead distance for a conveyor-serviced production station（CSPS） to improve the efficiency of operations.The optimal control process for CSPS is modeled by a semi-Markov decision process（SMDP）.Since the standard Q-learning is difficult to deal with the continuous variable optimal look-ahead control problem of CSPS directly,Cerebellar Model Articulation Controller（CMAC） for Q-values function approximation is combined with the online learning technology,and some online Q-learning and model-free online policy iteration algorithms are provided.Simulation results show that the proposed algorithms improve the learning speed and the precision of optimization.