The Study Of Reinforcement Learning Based Elevator Group Scheduling Technology

With the rapid development of modern urban high-rise buildings and intelligent building, it puts forward higher requirements on the efficient operation of the passenger and goods elevator group. It is one of the most important problems of the elevator group control to effectively improve the efficiency of elevator group scheduling, reduce operating costs, reduce elevator waiting and taking the ladder time and improve the distribution of elevator occupants, effectiveness and rationality.Based on different time of elevator operation rules, the author takes it as the research object to work during the upward lift centralized model, work during the elevator down centralized mode and work during the inter layer random traffic mode and idle mode. Then the author can find out operation rules, and make the scheduling system process information fast and accurately, realize online identification of the different traffic flow, and improve the response speed of scheduling through the establishment of model, analysis and identification.Elevator group scheduling control system project has some characteristics such as uncertainty and complexity. In order to handle possible unknown state events and make a timely response, the author uses machine learning in reinforcement learning algorithm to solve the problem in this paper. Based on Markov decision theory, reinforcement learning is an unsupervised learning method through trial and error and continuous self-improvement. Through reinforcement learning, the author can find the best solution to the current problems in the control system when having enough learning time and enough search range. In addition, the author also introduces the convergence of random search strategy based on simulated annealing algorithm and local optimal CMAC neural approximation based on the network to improve the scheduling search strategies and search in this paper. Dispatching it as an organic whole is based on the effective combination of Q-learning and other learning methods.The author firstly analyzes the status of the elevator group control scheduling in this paper, describes the existence of elevator group scheduling problem and then compare different elevator group scheduling algorithm. Moreover, In turn, the author introduces the reinforcement learning theory and advantages of the method, define the modules of the system and realize the function and corresponding work flow based on reinforcement learning of elevator group control scheduling mode. At last, the author verifies the feasibility and validity of the proposed method compared with other traffic model by the elevator group control virtual software and the simulation experiments.