User Incentive Based Bike-Sharing Dispatching Strategy

Abstract

AbstractAs a green and low-carbon transportation way, bike-sharing provides a lot of convenience in the daily traveling. However, after a period of usage, the distribution of shared bikes may not meet the traveling requirements of users. Shared bikes may converged in some areas while in other areas users have no bikes to ride. Therefore, the bike-sharing platform needs to dispatch the shared bike effectively to improve the user service ratio. Instead of using trucks to dispatch bikes, one possible way is to incentivize users to return shared bikes to the desired destination by subsidizing certain monetary, while ensuring users reach their original destinations within walking distance. However, the platform usually has a limited budget to incentivize users. In this paper, we design a bike-sharing dispatching strategy by incentivizing users to improve the user service ratio by taking into account the budget constraint, users’ maximum walking distance, users’ riding demands and dynamic changes of the distribution of shared bikes. The dispatching strategy consists of budget allocation and task allocation algorithms. In the budget allocation algorithm, we first predict the user’s riding demand based on LSTM, so as to generate dispatching tasks, then we model the allocation of budgets for each time step as a Markov decision process, and then design a budget allocation algorithm based on the deep deterministic strategy gradient algorithm. In the task allocation algorithm, due to the budget constraint that makes it impossible to use the mainstream bipartite graph matching algorithm, we choose to use the greedy matching algorithm for the task allocation. Finally, we run experiments based on the Mobike dataset to evaluate our strategy against greedy budget algorithm, unlimited budget algorithm, and the truck hauling algorithm. The experimental results show that our shared bike dispatching strategy with user incentive can outperform other benchmark dispatching strategies and maximize the long-term user service rate with limited budgets.KeywordsBike-sharing dispatchingDemand predictionUser incentiveDeep reinforcement learning