Stochastic agent-based models optimization applied to the problem of rebalancing bike-share systems

This paper presents an agent-based model employing a stochastic optimization search that attempts to find an optimal solution to the online bicycle rebalancing problem for general bike-sharing systems. The algorithm receives the initial and final global state configuration of the system. The main objective of the study is to find the minimum cost path from the initial to the final state. Each agent of the model has four behavioral options that search the optimal configuration; at each iteration, it selects one of these options based on random thresholds and shares the temporary solution found with neighboring agents to improve their search process. The algorithm presents a high exploratory behavior of the search space, which helps to find an approximation away from the local optimal configuration. Additionally, the exchanges between agents allow a consensus on the solutions found. The algorithm has been tested with two different generated configurations using as a basis a real dataset extracted from a functional bike-sharing system collected in 2019. The results show a positive evolution originating from the emerging effect of stochastic selection and interaction between agents.