Customized Bus Route Optimization Based on Reinforcement Learning

The customized bus plays an important role in catering the special needs of bus travel. The optimization method of customized bus route planning was studied considering the time windows of passengers. The arrival time of customized bus at stops were converted to equivalent bus travel distances. Taking the minimum weighted cost of the bus operator, passengers travel, and exhaust pollution, a customized bus route planning model based on the time windows of passengers was developed. The ant colony algorithm was designed to obtain a feasible solution of the planning model. Then, Q-learning algorithm based on action model approximation was designed to optimize the feasible solution. The environment model was mapped into a two-dimensional raster model, the action space was set, and the reward and punishment functions were designed. The results show that the optimized bus line can screen out the service stops, and the comprehensive cost is reduced by 20.2%.