To deal with the uncertainty of the elevator traffic flow and the destination of passengers, overcome the computing bottleneck existed in collective EGCS, an elevator group control scheduling method is proposed based on destination registration and multi-agent coordination. The EGCS was firstly designed and the multi-objective evaluation function of destination registration was demonstrated. Then, the online update of elevator traffic pattern recognition based on fuzzy logic was proposed, to update different traffic modes corresponding to the weights of the control objectives in real-time. At last, the elevator car and group controller can be mapped to different agents. The distributed elevator group control scheduling is to implement a multi-objective optimization by various agents, with a coordination of minimum average waiting time (AWT), average riding time (ART), congestion degree (CD) and energy consumption EC).