Research On Elevator Traffic Flow Prediction Based On Improved Neural Network

With the acceleration of urbanization process and the rapid development of high-rise buildings,it's becoming more and more important to improve vertical traffic efficiency.As the basis of elevator configuration planning and group control scheduling,elevator traffic flow prediction is the premise of solving the vertical traffic problems.Up to now,there have been various traditional and intelligent methods to predict traffic flow.However,a single algorithm can't meet the requirements of modern intelligent traffic system and the prediction trend is the combination of traditional and intelligent methods now.This paper combines time series method with neural network and establishes an elevator traffic flow prediction model based on BP neural network.Then ant colony optimization(ACO)algorithm is proposed to improve BP neural network and simulation results show the validity and reliability of ACO-BP neural network used in elevator traffic flow prediction.Firstly,the traffic flow data of passengers entering and leaving an office building are collected and normalized to form time series.The periodicity of the time series show the prediction possibility and the randomness and chaos show the difficulty.The traffic flow in one day is divided into four periods: morning up peek,noon up/down peek,afternoon down peek and random conventional period which are analyzed respectively to conduct the selection and modeling of prediction method.Secondly,basic knowledge of neural network architecture and learning rule are summarized.The BP neural network prediction model is built,which is mainly introduced from the aspects of node determination in each layer,sample library generation and BP rule.Traffic flow prediction is the use of BP neural network model to forecast data at the future moments according to data at the historic moments.Simulation results show that the traditional BP model can track the changing trend of elevator traffic flow with small time lag and low precision.Besides,it is easy to get trapped in a local minimum value.Finally,the math model of ant colony algorithm is presented under the background of TSP.In the face of large scale problems,original ant colony algorithm may have a slow convergence speed,consume a long time or present premature convergence.Thus,rank-based elitist ant strategy is proposed.After introducing the concrete method by which ACO algorithm is used to improve BP neural network,the ACO-BP model is built to simulate elevator traffic flow prediction.Compared with the traditional BP network,ACO-BP neural network has advantages in training process and prediction accuracy.Most importantly,it can search for the optimal solution globally without the influence of local minimum value.ACO-BP neural network performs well in elevator traffic flow prediction.