Research On Demand Forecasting Of Shared Bicycle Users Based On Improved LSTM Algorithm

Shared bicycles have developed rapidly in recent years for its advantages such as convenience,green environmental protection,and alleviating traffic pressure,but there are problems such as unbalanced supply and demand of shared bicycles and unreasonable delivery by operators currently.In addition to the unreasonable investment in the capital market behind these problems,the operator’s inaccurate prediction of the spatio-temporal distribution of shared bicycle demand and the subsequent unreasonable resource scheduling have also caused significant negative impacts.Therefore,operators need refined data analysis,accurate spatio-temporal demand forecasting,timely scheduling schemes and reasonable management strategies to achieve the sustainable development of shared bicycle systems.The thesis aims to analyze the spatio-temporal distribution characteristics of shared bicycle travel based on the historical order data of shared bicycles,and use deep learning algorithms to predict the demand of shared bicycles with different spatial granularity in different time periods,so as to provide support for subsequent vehicle scheduling and management strategies.The research content of the thesis is as follows:Firstly,the historical order data of shared bicycles in Shanghai in August 2016,and the travel characteristics of shared bicycle users from the combined dimensions of time,space and time are analyzed in the thesis.Secondly,combined with the weather data and Shanghai POI data,the Spearman’s rank correlation coefficient is used to explore the influencing factors of shared bicycle travel,so the important influencing factors are determined.Then,the shared bicycle demand prediction models: WT-LSTM and WT-conv LSTM are constructed in the thesis.The wavelet transform threshold method is used to denoise the demand data,and long short-term memory neural network and convolutional long short-term memory neural network are used to predict demand.Finally,the two prediction models constructed are experimented with and the prediction results are analyzed and evaluated.The demand of shared bicycle users on weekdays and non-weekdays under different spatial granularities in downtown Shanghai are predicted.The spatial aggregation and prediction accuracy of different spatial granularity are evaluated by PAI and PEI indicators,and the prediction accuracy of the model is evaluated by indicators such as mean square error.By entering different dimensional data,it finds that the WT-LSTM model added geospatial information and surrounding raster demand is more accurate than the forecast accuracy of the weather factors only,indicating that the demand for shared bicycles has spatio-temporal characteristics.The overall results of the WT-conv LSTM model predicting spatial granularity of 1kilometer is found to be good by comparing the prediction results of the three stages of morning,middle and evening on weekdays and non-working days.It finds that the forecast results of WTconv LSTM are affected by the number of demand,and the forecast effect is the best during the demand-intensive period.Compared with the forecast results of other forecasting models,WTconv LSTM has the highest accuracy,and its prediction error is MSE=0.00286,RMSE=0.0535,and MAE=0.0264.WT-conv LSTM improves the spatio-temporal prediction accuracy of shared bicycle user demand to a certain extent,and lays a research foundation for the scheduling and management strategy of shared bicycle resources.