Parallel Computing Of Route Planning And Vehicle Updating In Microscopic Traffic Simulation

Microscopic Traffic Simulation (MTS) simulates the running of the real traffic flow realistic throughdescribing the model of each vehicle’s behavior, and it’s convenient for traffic management, control,planning and forecasting. With the expanding of the city scale, the increasing of the vehicles, and thecoupling relationship between the traffic signal control and the traffic flow, the road network scale and thenumber of the vehicles in traffic simulation system are constantly increasing, and the requirement ofsimulation effect, the real-time performance and the analysis function for the simulation software isincreasing. Using traditional sequential MTS is difficult to meet the requirements of real-time simulation inlarge scale road network. Parallel computing is an effective way to improve the performance of computing,and parallel MTS uses multiple computing resources to complete the simulation collaboratively, whichimprove the efficiency of MTS.The paper analyzes the parallelizability and the basic parallelization approach of MTS based onsummarizing the typical parallel MTS methods, and the paper designs the multi-computers working modeof master-slave combining with the MPI feature, which can implement the distributed parallel computingplatform via connecting multiple computers by network. The main achievements of the work include:(1) In the heterogeneous system of GPU collaborative with CPU, the paper uses CUDA to realize theparallel computing of path planning, and the experiments under the large-scale road network data show thatthe method can effectively improve the efficiency of path planning.(2) Combining with the representation of the three layers structure of road network, this paper usesdomain decomposition strategy to partition task, then allocates the road network averagely to multiplecomputers by the serial number of the roads and intersections, and completes the simulation coordinatelywith these computers.(3) The paper uses MPI message passing mechanism to implement communication between thedistributed multi-machines that in master-slave mode: the simulation nodes use multi-thread technique toimplement message sending and receiving, and the master node collects all information of vehicles at eachstep of the simulation.(4) The paper designs the simulation experiment under different road network and different vehiclenumber, and analyzes the main factors affecting the speedup of distributed parallel MTS.