Exploiting Taxi Demand Hotspots And Guiding For Taxi Cruise Based On Vehicular Big Data Analytics

With the urban population growth and economic development,urban traffic is under increasing pressure.Taxis play an important role in urban transportation,bring convenience to the public,but also bring about traffic congestion,waste of energy and environmental pollution.There is high empty rate of taxi while there is a problem of calling taxi,so improving the efficiency of the taxi is a key to ease urban traffic pressure and resource conservation.With the development of information technology,many of the taxis are equipped with GPS(Global Position System)equipment,these devices send their own real-time information to the data center every few seconds including vehicle location,speed,passenger status and so on.How to use these huge operational data to dig out useful information to improve the efficiency of taxi is a popular research field,and also the focus of intelligent transportation and large-scale data mining.Based on the above analysis,In this paper we process data using some large-data processing platform such as Hadoop or spark and design the entire process framework which starts with raw data and ends with advice about cruise direction to the taxi driver.we make use of data to reduce the taxi empty rate and improve the efficiency of taxi.The main work of this paper includes:1.Using the existing data to analyze the characteristics of urban taxi demand,we found that the demand size of the same area in the time dimension is very similar,but the difference of model in different areas is quite big.In addition,it was found that even in the case of different number of sampling vehicles,the sensing result about traffic condition of the same area was not greatly different.2.Through clustering on and off the taxi point to perceive the city hot spots.An improved version of the DBSCAN clustering algorithm(GD-DBSCAN)is proposed to improve the clustering speed of the algorithm by using the partition and KD-Tree index.The performance of the algorithm is improved by about 10% and we reform original serial algorithm to support parallel processing.3.This paper proposes the idea of using pick-up rate to evaluate the probability of hotspots in the unit time,not only digging hot spots,but also assessing the heat size of different hot spots.Deep learning algorithm called Long short term memory(LSTM)is applied to the forecasting hotspots,and the deep network model is established to improve the accuracy of prediction in the hotspots.Based on the forecast of the next few moments' pick-up rates of different hotspots,we help taxi driver to choose the best cruise direction and route.