Research On Taxi Demand Prediction And Hot Spot Ride-sharing Algorithm Based On Big Traffic Data

With the rapid development of the economy,the speed of the city planning and construction is much lower than the growth rate of the vehicle,which accordingly causes road congestion.Due to the limitation of ride location and time for public transportation such as the bus and the subway,passengers who value the travel convenience and flexibility will tend to travel by taxi.However,in peak hours and road sections,passengers often have difficulty in getting a taxi.The main reason for this phenomenon is the unbalanced supply and demand relationship between taxis and passengers,and this problem cannot be solved fundamentally by only increasing the number of taxis.The popularity of the Internet and GPS technology provides an opportunity for the development of taxi platforms and also facilitates the collection of taxi data.Using association rule mining algorithm can discover the passengers' traveling rules and realize the taxi demand prediction.It can reduce the no-load rate of taxis and help passengers find taxis easily,through which the supply and demand relationship can also be balanced.This process can be regarded as the non-carpooling dispatching of taxis.At the same time,ride-sharing as a new trip method has been gradually accepted by the public.To a certain extent,it can reduce the traffic congestion and the air pollution,and it can help passengers to save on taxi fares.The research on carpooling scheduling system is mainly to solve the problem of dynamic matching and scheduling between passengers and taxis.Compared with non-carpooling scheduling,it is more complicated for real-time processing of request data,and more constraints are involved in the matching process.For the taxi demand prediction problem,the association rule mining can be used to analyze the temporal and spatial relationship between passengers' getting on and off behavior.For association rule mining the most time-consuming process is the frequent itemset mining.The processing speed should be increased as much as possible when dealing with the massive traffic data.According to the computing characteristics of the CUDA framework,the frequent itemset mining algorithm can be applied to the GPU acceleration through the graph-join method and the bitmap dynamic queue,which can boost the performance greatly.Through the empirical comparisons on a large Accident dataset,our GPU accelerated implementation can achieve 63 x speedup ratio compared to the classic Goethals version of Apriori algorithm.And compared to the classic Bodon version,our GPU accelerated implementation on GeForce GTX 1080 graphic processor can achieve 19 x speedup ratio.Our GPU accelerated implementation performs better when dealing with a longer transaction dataset.For the carpooling scheduling system,we summarize the hotspots of each time period from trip record data.The accuracy of geographical grid division is improved in the hot spot area,and the range of taxi candidate sets will be more reasonable.The speed attenuation zone is introduced into the hot spot area,and based on the speed attenuation zone,we describe the travel time estimation method.The optimized ride-sharing system is verified by 203,047 taxi requests generated by the Taxi Request Generator simulator according to the taxi history requests in Beijing Chaoyang District.The experimental result and analysis show that on the premise of not affecting the average user satisfaction and not significantly reducing the system service rate,compared with the ride-sharing system without the taxi candidate set search strategy,this optimized ride-sharing system reduced the calculation by 70.73% in average.