The problem of elevator group scheduling is formulated by the framework of the Markov Decision Process (MDP), and then the elements in the model of a reinforcement learning algorithm are defined. When reinforcement learning is applied, the stochastic action-selected policy and feed-forward neural network are used to handle the problems of exploration and generalization of value function respectively, which are integrated into the value iteration algorithm, called "Q-learning," to build up the whole algorithm for elevator group scheduling. The simulation results demonstrate the good learning ability, good performance and the adaptability for different traffic flows of algorithm scheduling.
展开▼
