Spatiotemporal Clustering And LSTM Based Prediction Model Of Bicycle Sharing

Bicycle sharing has been favored in recent years as an auxiliary means of urban public transportation due to its advantages of convenience and environmental protection.Nonetheless,bicycle sharing is extremely prone to unbalanced resource allocation in practical use,that is,while some regions are in short supply,others may be in oversupply.Therefore,it is of great significance to discuss the resource allocation of shared bicycle in order to reduce the operation cost and improve the user satisfaction.Aiming at the lack of existing research on the problem of mining the complex spatiotemporal relationships of bicycle sharing systems,we explored the method of dividing bicycle parking stations based on spatiotemporal clustering and the related bicycle demand forecasting model.(1)Aiming at the parking station selection for Pileless shared bicycles,a spatiotemporal clustering method combining ST-DBSCAN and K-means was proposed to reasonably divide bicycle parking stations.Compared with the traditional methods,the proposed method can be superior in determining the capacity of the bicycle parking station,the initial position of the station centroid,and the number of stations.By analyzing the bicycle POI(Point of interest-Destination)matrix requirement distribution based on the bicycle trip dataset,the rationality and effectiveness of the proposed method applied to the division of shared bicycle parking area were verified.(2)Aiming at the time series characteristics of travel rules,a demand forecasting model for shared bicycles based on long-term and short-term memory networks was established,which realized the demand forecasting of shared bicycles that combined the time and space dimensions.This model takes into account various factors such as working and non-working days,weather conditions,time periods,and geographic information of each parking station.The experimental results showed that the minimum root mean square error of the proposed model was 0.060,and the prediction accuracy was 0.900;under the same conditions,the forecasting effect of the demand for bicycle sharing was better than the prediction model constructed by the multi-layer feedforward neural network and recurrent neural network based on error back propagation training.The proposed model has certain value in assisting operators to solve the resource scheduling and resource allocation problems of shared bicycles.