Shared Cycling Demand Prediction Combined With Urban Computing And Spatiotemporal Residual Network

In recent years,with the rapid development of social economy,traffic congestion,traffic pollution,traffic accidents and other problems associated with the continuous growth of motor vehicle ownership have attracted wide attention.The strategy of building a powerful transportation country puts forward new requirements of "quality and quantity preservation" for the development of transportation,among which the sustainable development concepts such as carbon neutrality and carbon peaking have gained popular support,and the sharing transportation industry has emerged at the historic moment.As one of the main forces of urban shared travel,shared bikes have advantages such as resource sharing,green and low-carbon,healthy and fashionable.However,the rapid development and improper regulation of the bike-sharing system has led to the encroachment of shared bikes on public roads and space resources,affecting the urban chronic traffic order and bringing hidden dangers to traffic safety management.The demand forecast of shared bikes is the cornerstone of scientific and reasonable planning of shared bike resources.Therefore,it is necessary to study how to accurately forecast the demand for shared bikes,and build a real-time,efficient and accurate demand forecast system for shared bikes,so as to provide strong support for parking planning,resource allocation and reset,and route guidance of shared bikes.It is of great significance in improving residents’ slow travel environment,enhancing the competitiveness of shared travel,promoting related enterprises to reduce costs and increase efficiency,optimizing the resource allocation of urban slow traffic system,alleviating traffic congestion,saving energy and reducing emissions,etc.In view of the above background,this study was carried out in order to deeply explore the complex space-time rules and characteristics of shared bike travel,reveal the influence of urban factors on shared bike travel demand,and improve the prediction accuracy of travel demand.From shenzhen city of open government data platform for sharing large orders by a single data set,data and urban factors based on python development environment visualization analysis shared bike travel correlation between time and space distribution rule and its influencing factors.Based on the urban calculation theory,a spatial-temporal attention residual network prediction model(USTARN)was constructed for the travel demand of shared bicycles,considering the epidemic situation,weather conditions,minimum temperature,maximum temperature,wind speed,working days and non-motorized lane length.First,USTARN captures the spatio-temporal dependence of single vehicle flow by partitioning the occurrence,attraction or OD data of shared bikes through spatial division and time series segmentation of different duration characteristics.Then,combined with the attention mechanism,it conducts deep residual learning.Finally,it adjusts the prediction results according to the prediction results of urban factors to improve the prediction performance of the model.The bike-sharing data set and urban factor data set was divided into a training set,verification set and test set according to 7:1:2,and the training prediction,adaptive adjustment of model parameters and comparison experiment of model validation were carried out respectively.Studies have shown that USTARN model sharing of bicycle travel can in a shorter computing time,accurate to solve the share of each district bike travel demand,and the more traditional S-LSTM,BiLSTM,CNN model error is lower and has better prediction performance.The USTARN model fully reflects the influence of time,space,epidemic,weather,temperature,wind speed,non-motorized lane length and other factors on the travel demand of shared bikes.