Data-driven route guidance under the framework of model predictive control

Real-time traffic assignment for route guidance is put under the framework of model predictive control, which optimizes the routes based on the real-time feedback and prediction information of road network. In this framework, particle filter is utilized to estimate the statistic distribution of traffic flow of links without detection sensors based on the position and speed information of navigated vehicles on those links and the prior information of traffic flow of links with detection sensors. The chance constrains and Bayes-based route prediction are incorporated into the optimization model so that the stochastic characteristics of traffic needs, propagation and driver´s decision-making behavior can be compensated in the route optimization. To check the chance constraints, the min-max characteristic points are used to fit the curve of stochastic traffic propagation process with stochastic needs to avoid the exponential increase of combination calculation. The genetic algorithm is utilized for the optimization with feasible-direction-search crossover and mutation to improve the evolution efficiency, combined with the traffic simulation in the mean sense with the compensation of stochastic parts of traffic flow data to evaluate the performance of real-time traffic assignment. The simulation results demonstrate the effectiveness of traffic navigation predictive control.