Elevator traffic flow prediction can provide essential traffic data information for elevator configuration and group control scheduling. Considering that the established elevator traffic flow prediction model based on BP neural network is sensitive to initial value and easy to get trapped in some local minimum value in process of neural network training

ant colony optimization（ ACO） algorithm with global optimization capability is proposed to optimize BP neural network. Besides

elitist ants and rank-based strategy are used to improve the basic ACO algorithm. The real elevator traffic flow data of an office building in Tianjin are collected to analyze and simulate these models. Simulation results show that the prediction effect of ACO-BP neural network is far superior to traditional BP neural network and ACO-BP is suitable for the elevator traffic flow prediction system.