Research And Application Of Demand Prediction And Optimal Rebalancing Algorithm In Bike-Sharing System

The emergence of public sharing-bike provides a convenient service for people to commute and daily travel,and also brings new vitality to urban transportation services.In actual use,due to some inevitable reasons,it often happens that there are no bikes or docks available at the stations,and the bike-sharing system needs to reallocate and rebalance bikes resources between stations frequently.The shortage situation that often occurs at the station makes the station cluster more able to reflect people’s real demand for bikes.Therefore,it is of great significance to accurately predict the bike demand of the station clusters in different periods and to meet the bike demand of the station clusters to make full use of bike resources in bike-sharing system.However,the reflection of the real bike demand in the station clusters and the value of the station clusters in the bike resources rebalancing are often ignored.This paper studies the above issues,and the specific research contents are as follows.This paper will solve the rebalancing and scheduling problem for docked bike-sharing system in two phases,which are station clusters mining and demand prediction phase and static rebalance scheduling path planning phase.In the first phase,the bike usage pattern of the stations to be mined by processing the riding records,and a dynamic clustering algorithm named Location-aware Hierarchal Clustering algorithm(LHC)for stations clustering by considering both bike usage demand and geographical location between the stations is proposed.This paper comprehensively analyzes the influence of a series of potential features in multi-source data on bike demand,then fully models the extracted relevant features and proposes a XGBoost-based regression model for temporal modeling to predict the cluster-level rental and return demand.In the second phase,this paper proposes a historical time window K-Nearest Neighbor and Redistribution(HTKR)algorithm to estimate the bike demand of each station in the clusters at different time periods,so as to the greatest extent meet the bike prediction demand of the station clusters.Then through studying and analyzing the advantages and disadvantages of ant colony algorithm and derivative ant colony system,combined with chaos theory,an ant colony optimization algorithm is proposed to solve the static rebalance scheduling path planning problem.Sufficient experiments on real-world datasets of the Citi Bike system in New York are conducted to verify and evaluate the effectiveness of the methods proposed in this paper.The experimental results confirm that compared with the comparison method,the LHC-XGBoost has a more accurate prediction results for the bike demand prediction of the station clusters.At the same time,in the static bike rebalance scheduling problem,HTKR-ACO can satisfy more than 88% of the predicted bike demand of the clusters after rebalancing bike resources,and can find the scheduling path with the least scheduling cost at a faster convergence speed than the comparison method.It can effectively plan a reasonable scheduling path for the system.Finally,based on the above two stages of work,we designed and implemented a practical application of bike-sharing rebalancing system using PyQT5 to guide the practice of bike resources rebalancing and scheduling path planning,and improve the availability of bike sharing service.