Bike-Sharing Travel Demand Forecasting And Scheduling Optimization

As a low-carbon,convenient and low-cost way to travel,bicycle sharing is gradually becoming an indispensable way to travel short distances in people’s daily lives.With the increase of users and the development of technology,the ability to improve the operation level becomes the key to competition among bike-sharing operation companies.The operation of shared bicycles needs to be more scientific,intelligent and precise.How to predict the demand of shared bicycle and make more efficient scheduling based on the prediction results combined with the existing actual situation is directly related to the operation cost of shared bicycle operation enterprises and the satisfaction of shared bicycle users.In this paper,we propose a set of solutions to improve the accuracy of bike-sharing demand prediction and to achieve more efficient and reasonable scheduling.In summary,this paper mainly does the following research(1)A combined prediction model based on complete ensemble empirical mode decomposition with adaptive noise,wavelet threshold denoising,bi-directional long and shortterm memory network,and gated recurrent unit network is proposed for the bike-sharing travel demand prediction problem.The data input to the model is firstly noise reduced to improve the intrinsic correlation of the data,and then the data is decomposed into multiple component data by empirical modalities,and further high and low frequency identification and reconstruction of individual component data are performed to strengthen the intrinsic features of the data,and then the neural network is selected for prediction according to the data with different features.The real data provided by the public dataset are used,and the data are firstly cleaned and preprocessed to eliminate invalid data and then aggregated into multiple travel data according to the time granularity of one hour,while some latitude and longitude data are selected and input into the model separately.The test set and training set are divided according to 30% and 70%,and compared with several other neural network-based combined prediction models.The results show that the performance of the proposed model in this paper is quite competitive in accurately predicting long sequence information at different scales.Eventually,the spatial and temporal distribution of bicycle usage demand for the next hour is predicted by the model based on the tail data of real data.(2)The spatio-temporal distribution information of demand obtained from the prediction results is first operated by DBSCAN clustering to identify outlier demand points;and after recording and eliminating the outlier demand points,K-means clustering is performed on the remaining data to cluster the remaining data into multiple clusters with different demands.The cluster center of each cluster is marked as the bicycle dispatching station responsible for the demand in that area.The final result yields multiple shared bike dispatching sites divided based on user demand for bikes and multiple outlier demand sites that need to be considered separately in the dispatching process.(3)Based on the results of clustering,multiple regular scheduling sites for shared bicycles are identified and outlier demand points are abstracted into special sites with special demand and special locations.The regular sites and special sites are incorporated into the unified index management system,while the vacant bicycle parking situation of the previous time period is considered comprehensively.An economically induced user self-dispatching strategy is introduced,a bi-objective optimization model with the objective of minimizing enterprise cost and maximizing site satisfaction is established,and the NSGA-II algorithm is designed to solve the problem,and the Pareto non-inferior solution set for the problem is finally obtained.