Research And Design Of Elevator Group Control System Based On Multi-agent

With the development of modern city, more and more skyscrapers are built in the city today, and the traffic in the skyscrapers become more complex. One elevator can't satisfy the needs of all passengers. It is indispensable to transport passengers with elevator group composed of several lifts. This kind of system, which organizes many lifts together and dispatches them in reasonable way, is elevator group control system (EGCS). Accordingly, the elevator group control method has already become the focus of research. Contraposing some disadvantages of current methods, this paper apply multi-agent theory to EGCS, so that the mass computing of former centralized control system can be separated. Simultaneously it uses reinforcement learning method to make EGCS adaptive to changefully traffic flow and environment. Accordingly the running efficiency of EGCS is increased.Firstly, the thesis reviews the development and current status of EGCS, as well as correlative control method, and introduces advantages and disadvantages of varieties of EGCS. The thesis analyses the systematic characteristic, control method, performance evaluation index and traffic flow of EGCS, and then sets up analyzing models of traffic flow. The origination density vector and origination-destination matrix are obtained through ratiocination. The thesis simulates the traffic flow model and gets the traffic data.Then this thesis designs the EGCS based on multi-agent theory. Every elevator is designed as an elevator agent, which can not only control its cage's moving, but also make independent decision to select a call suitable to response for itself. At the same time every elevator agent learns from the results of call responding through reinforcement learning, so that the whole elevator system can adapt well to diverse traffic flow and obtain good performance. When there is conflict between calls selected by elevator agent, a coordinating agent designed in the thesis will resolve this conflict according to coordinating mechanism. The thesis designs the system frame of EGCS based on multi-agent, the structure of elevator agent and coordinating agent, the decision-making and reinforcement learning method of elevator agent, and the coordinating mechanism of coordinating agent.Finally the thesis constructs an elevator group control simulation system basd on multi-agent using object-oriented programming language C++. Simulation results represent that the EGCS based on multi-agent can abate the mass and centralized computing of former centralized EGCS, and improve computing efficiency consequently. Furthermore it adapts well to traffic flow transformation and environment break.