Optimization Of Planning And Dispatching Approach For Bike-Sharing System

In recent years,by the era of the sharing economy and mobile Internet,bike-sharing has developed rapidly all over the world.The emergence of bike-sharing has solved the "last mile" travel problem to a certain extent,and built a bridge between traditional travel modes.With the development of high-capacity rechargeable battery technology,motorcycle-sharing are emerging in the market,which provides a more convenient and labor-saving way for medium and short distance travel.However,with its rapid development,there have been many problems,such as the accumulation of bicycles,unbalanced supply and demand,and dead batteries,which not only seriously affect the user experience,but also bring huge challenges to urban management.Therefore,this thesis studies the planning and dispatching of bike-sharing system from three aspects.(1)For the phenomenon of asymmetric supply and demand,we propose a Spatio-Temporal Mixed Integer Programming(STMIP)with Flow-graphed Community Discovery(FCD)approach to rebalancing the system based on trucks.First,we propose the FCD algorithm to detect stations and station communities.Each station community is inner-balance and inter-independent.Then,aiming at maximizing the total profit of the system,we design a STMIP model to solve the multi-trucks routing problem in each community holistically.The STMIP with FCD approach not only decomposes the centralized control system into a distributed multicommunities system,which heavily reduces the complexity of the multi-trucks routing problem,but also considers the routing and velocity of capacitated trucks,which greatly improves the effectiveness of dispatching.(2)For the rebalancing problem during the peak period when the distribution of supply and demand changes rapidly,we propose a Multi-Reward Cascaded Deep Q-networks combined with Bottom-up Strategy(BS-MRCDQN)dispatching algorithm.And the dispatching is decomposed into destination station selection and loading level selection.In this algorithm,the Multi-Reward Cascaded DQN(MRCDQN)dispatching module is established in the form of multi-agent,and used to obtain the combined optimal strategy of two sub-tasks.When a truck completes the previous dispatch,its state is updated and a new task is generated immediately without waiting for other trucks to complete.Compared with the overall planning,this approach can capture the dynamic changes of the system more timely and improve the dispatching efficiency.In order to reduce the action space,we limit the alternative destination station to adjacent areas,so that the same supply and demand situation will tend to stop the trucks in place.The Bottom-up Strategy(BS)can move long-standing trucks out of the area to restore their rebalancing capabilities in time.(3)For the problem of the battery replacement and supply-demand imbalance in the motorcycle-sharing system,we consider a truck that can load motorcycles and rechargeable batteries simultaneously,and design a Multi-Objective Mixed Integer Programming(MOMIP)model to generate trucks’ routing.The model includes two objective functions:maximizing the total profit and maximizing the number of available motorcycles.We allow the truck carry out the motorcycles and replacing low-power batteries simultaneously at the visited station.By this way,the trucks’ routing is the optimal routing which explores the trade-off between the system imbalance and reducing the number of unavailable motorcycles.It can not only effectively improve the user service level,but also reduce the moving distance of trucks when the two problems are solved separately.