| With the rapid growth of motor vehicles in cities,traffic problems such as air pollution,traffic congestion,and traveling inconvenience have become increasingly prominent.The development of public transportation systems has become a consensus.As an indispensable part of urban public transportation systems,bike-sharing can improve the utilization of road resources,alleviate traffic congestion,and effectively solve the "last mile" problem.It is a green travel mode that fits the low-carbon development requirement of current society.Currently,the bike-sharing systems operated worldwide can be divided into two categories: docked bikesharing and dockless.Dockless bike-sharing system,as a new service mode developed by the integration of Internet and bike-sharing,has developed rapidly in China in recent years.Compared with traditional docked bike-sharing,dockless bike-sharing companies,which are still under development in the early stage,have relatively less experience in practical operating,resulting in the lack of theoretical guidance for operation and management,and restricting the sustainable development of the entire system.In this paper,the spatiotemporal characteristics of dockless bike-sharing are analyzed based on multi-source data.A short-term demand prediction model based on deep learning is further constructed.Then,a rebalancing strategy involving users' participation through the price incentive mechanism is proposed from a micro perspective.Optimization models of rebalancing in both static and dynamic dockless bikesharing systems are proposed from a macro perspective,followed by corresponding solving algorithms designed.Numerical experiments on the models above are conducted based on the historical bike-sharing trip data of Nanjing,China.Firstly,based on format optimization and data cleaning of multi-source data,spatial attributes of multi-source data are integrated through the GIS platform.Using historical data of dockless bike-sharing as input,centroids of the virtual stations are obtained with the spatial clustering algorithm.Voronoi Diagram is further created using the centroids as the control point and the virtual stations of the dockless bike-sharing system are thus generated.Data mining approaches,together with spatial analysis tools are applied to analyze the characteristics and evolution of dockless bike-sharing temporally and spatially.The results show that the performance of the K-means algorithm outperforms the others when generating virtual stations.The cycling time and distance share similar distribution on weekdays and weekends,with a significant spatiotemporal imbalance and tidal phenomenon in traveling demand of dockless bike-sharing.Secondly,a spatiotemporal graph convolutional neural network integrated into the attention mechanism is proposed to predict the short-term demand for dockless bike-sharing.The long short-term memory(LSTM)network and the graph convolutional neural(GCN)network are integrated to extract the temporal and spatial characteristics of dockless bikesharing demand.The attention mechanism is introduced to find the internal relationship between the input sequence characteristics to improve the accuracy of the prediction model.The experimental results show that the prediction results of the spatiotemporal graph convolutional neural network model with attention mechanism(GATGCLSTM)are superior to the baseline models under different intervals.The prediction accuracy of the model can be further improved when external factors are introduced into the model.Predicted demand and the actual demand are visualized in a spatiotemporal perspective,which verifies the accuracy of the model.Thirdly,based on the short-term demand forecast results,a rebalancing strategy involving users' participation through the price incentive mechanism is proposed.Four rebalancing scenarios are set based on users' intended station for renting and returning the bikes,together with the state regarding the remaining bikes of the stations around reachable through walking.Subsequently,based on the urgency level of the station,incentive prices and incentive scale are calculated and sent to users.The decision on whether to participate in the rebalancing is made next by users through utility maximum theory.The results show that under the same incentive scale and customer participation probability,both Bikes Over Upper Bound(BOUB)and Bikes Below Lower Bound(BBLB)are lower in peak periods compared to off-peak periods.It is suggested that the incentive scale should be fixed from 1.30 to 1.40(average incentive price from 1.45 to 1.51 yuan per person)in peak periods,and 1.90 to 2.00(average incentive price from 2.11 to 2.14 yuan per person)in off-peak periods.Fourthly,considering the capacity limitations of virtual stations and user demands,a rebalancing demand model for dockless bike-sharing is proposed.Next,the station similarity matrix is constructed according to the rebalancing demand and distance among stations and the community discovery algorithm is used to divide sub-area for rebalancing.A static rebalancing route optimization model for dockless bike-sharing is proposed with the objective function consisting of two objects,namely minimizing rebalancing cost and the deviation between the target and actual rebalancing quantity of bikes.The immune genetic algorithm,which combines the selection memory mechanism of the immune algorithm and the traditional genetic algorithm,is designed to solve the model.Experiments based on real data are conducted to verify the proposed model above.The results show that using the immune genetic algorithm,the value of the objective function is better than that of the genetic algorithm.Rebalancing scheme provided by the model could meet 80.80% demand of all the stations.If the demand for every station is strictly satisfied,the rebalancing time will increase by 127.10% and the cost will increase by108.80%Finally,based on the dynamic complexity of the variation of users' demand,a dynamic rebalancing optimization model for dockless bike-sharing is proposed.The difference between renting and returning velocity of each station and the maximum capacity of each station is used to determine rebalancing demand.The concept of station importance is introduced,and the importance of each station is calculated based on the TOPSIS model.A dynamic optimization model,which minimizes rebalancing cost,as well as penalty cost on the deviation of the actual and target rebalancing quantity and maximizes the converted cost of users' satisfaction is proposed.A strategy based on rolling horizon strategy is further used to adjust the rebalancing scheme dynamically.The Artificial Bee Colony(ABC)algorithm considering the importance of the stations is designed to ensure that the important stations are served first,and the model is verified through numerical examples at last.The model results show that the objective function value and running speed of the ABC algorithm are significantly better than the Genetic Algorithm(GA).Specifically,the objective function value obtained by the ABC algorithm improved by 32.40% compared to GA,and the running time is lowered by 88.10%.Regarding the comparison between models considering the importance of stations or not,it is found that the model considering station importance can increase user satisfaction from 55.03% to 73.00%. |
PDF Full Text Request