结合加强学习的群控分区算法

运用加强学习算法解决电梯群控问题往往受限于算法收敛速度慢,很难扩展至具有更高楼层、更多电梯的群控系统.分割状态空间为几个区域,建立具有分割功能的加强学习系统是必要的.所提出的系统结构及其底层工作原理具有普遍意义,可用于解决大状态空间上的最优控制问题,开发了基于群控分区算法的分割模块,运行结果表明了此系统的优势.

Combine Reinforcement Learning with Zoning Algorithm

It is hard to apply reinforcement learning algorithms to solve elevator group control problem in a building with more floors and elevators. This is mainly because of low convergence speed of reinforcement learning algorithms. It is necessary to partition state space into several regions and establish a reinforcement learning system with partitioning function. The system framework and its inside performance principle have a general significance and can be applied to optimal control problem with large state space. A zoning algorithm based partitioning module was developed and the performance results show the advantage of such system.