Reinforcement learning approaches on elevator Group control problem

Elevators play a crucial role in the functionality of skyscrapers. As the number ofskyscrapers increases, so does the demand for efficient elevator systems. Traditionalsingle lift traffic control systems are inadequate for high-rise buildings, leading tothe adoption of multiple lift traffic control systems. These systems feature multipleshafts, each containing one or more elevator cars. Effective elevator group control(EGC) strategies are essential to ensure elevators work together to minimizepassenger waiting times.This thesis explores the application of reinforcement learning (RL) to enhance theperformance of elevator group control systems (EGCS). Given the NP-hard natureof scheduling elevator cars, where no known optimal solution exists, a learning-basedapproach offers a promising alternative to heuristic methods. This thesis proposestwo deep Q-network (DQN) techniques that enable the EGCS to dynamically assignpassenger calls to the most suitable elevator car, aiming to reduce passenger waitingtimes and improve overall system efficiency.