Automation in Construction ( IF 9.6 ) Pub Date : 2021-04-29 , DOI: 10.1016/j.autcon.2021.103737 Dongmin Lee , Minhoe Kim
Construction hoists at most building construction sites are manually controlled by human operators using their intuitions; as a result, unnecessary trips are often made when multiple hoists are operating simultaneously and/or when complicated hoist calls are requested. These trips increase a passenger's waiting time and lifting time, reducing the lifting performance of the hoists. To address this issue, the authors develop an autonomous hoist supported by a deep Q-network (DQN), a deep reinforcement learning method. The results show that the DQN algorithm can provide better control policy in complicated real-world hoist control situations than previous control algorithms, reducing the waiting time and lifting time of passengers by up to 86.7%. Such an automated hoist control system helps shorten the project schedule by increasing the lifting performance of multiple hoists at high-rise building construction sites.
中文翻译:
基于深度强化学习的高层建筑自动施工升降系统
大多数建筑工地的施工升降机是由操作人员凭直觉手动控制的;结果,当多个提升机同时运行和/或要求复杂的提升机呼叫时,经常会发生不必要的跳闸。这些行程增加了乘客的等待时间和起重时间,从而降低了提升机的起重性能。为了解决这个问题,作者开发了一种由深层Q网络(DQN)(一种深层强化学习方法)支持的自动葫芦。结果表明,与以往的控制算法相比,DQN算法在复杂的现实世界中的起重控制条件下可以提供更好的控制策略,减少乘客的等待时间和举升时间达86.7%。