Research On Bus Arrival Time Prediction Method Based On Sparse AVL Data

The reliability of bus arrival time is an important indicator for measuring the service level of the public transportation system.Improving the prediction accuracy of bus arrival time is of great significance for reducing passenger waiting time,increasing the proportion of urban residents’ bus trips,and alleviating urban traffic congestion.However,the spatiotemporal sparse distribution of bus trajectory data and the complex time-series change rule of the operating speed between bus stops have led to difficulties in the actual bus arrival time prediction.In order to solve the above problems,the following aspects are studied in this paper.First,given that the low sampling frequency of AVL data leads to the missing data of bus arrival time and reduces the estimating accuracy of the travelling speed between stations,a method based on low-frequency AVL data is proposed to estimate the travelling speed between stations.After the original AVL data is pretreated by map matching and cleaning of abnormal data,the bus arrival time under the three different operating conditions before arrival(travel in constant speed,slow down and stop by the station)is estimated according to the operating conditions of buses recorded in AVL data.And use the estimated arrival time to complete the estimation of bus travelling speed between bus stations.Next,given the impact of a spatiotemporal sparse distribution of the bus travelling speed sample data on the prediction of bus arrival time,an improved tensor reconstruction model is proposed to incorporate the temporal and spatial variations of travelling speed.Based on traditional Tucker reconstruction model,a constraint function is added to display the temporal and spatial variations of travelling speed.In this way,the estimated results of the improved tensor reconstruction model are more in line with the spatiotemporal variation rules of travelling speed,thus reducing the errors in the estimated results of sparse sample data.Finally,based on the demonstration that empirical mode decomposition(EMD)can effectively extract the time sequence characteristics of bus running speed,a bus arrival time prediction method based on EMD-LSTM model is proposed.On the one hand,given that the blending of various random and periodic timing characteristics in travelling speed leads to the incompetency of the prediction model in perceiving the sequential variation rules of speed,EMD is used to extract a range of timing characteristics mixed in the speed.In this way,the long-short term memory(LSTM)neural network model can fully perceive the underlying sequential variation rules in travelling speed,so as to improve the prediction accuracy of travelling speed.On the other hand,given that the bus arrival time cannot be updated in a timely manner due to the lack of real-time travelling speed caused by longer time headway,the timeliness in predicting bus arrival time is improved by estimating the travelling speed of buses in different sections during multiple time intervals.