强化学习A3C算法在电梯调度中的建模及应用

Modeling and application of reinforcement learning A3C in elevator scheduling algorithm

为让电梯调度算法在电梯电力能耗、用户乘梯体验和算法适应性方面具备更好表现,在目前主流的电梯调度算法基础之上,提出对调度环境、电梯行为和调度目标3个方面进行统一建模的基于强化学习A3C的电梯智能调度算法。让调度电梯在不断地和环境交互学习过程中逐渐学习得到最优电梯调度策略,与基于具体环境建模的相关电梯调度算法进行对比实验,基于A3C的调度算法具有建模简单规范、适应性强和控制目标多样的优势,对比A3C算法与部分强化学习算法在电梯调度中的优劣,实验结果表明,A3C算法具备较好的调度性能。

To make the elevator scheduling algorithm have better performance in the aspects of elevator power consumption,user experience and algorithm adaptability,based on some elevator scheduling algorithms,the elevator intelligent scheduling algorithm based on reinforcement learning A3C was constructed,which unified the three aspects of scheduling environment,elevator behavior and scheduling objectives.The elevator learned the optimal scheduling strategy in the process of continuous interactive learning with the environment.Compared with some elevator scheduling algorithms,the scheduling algorithm of reinforcement learning modeling has the advantages of simple modeling and high scheduling efficiency.At the same time,the experiment explores that A3C algorithm has better scheduling performance than some reinforcement learning algorithms.