Urban Shared Bicycle Travel Analysis And Regional Demand Forecast Based On Data Mining

Sharing bicycles as an emerging “Internet + Transportation” mode of travel has alleviated the city's noise pollution and traffic congestion.Moreover,with its advantages of flexibility,low carbon,environmental protection,economical energy saving and good accessibility,it has gradually become the preferred method to solve the “last mile” travel of the city.However,its unique characteristics have brought many problems to users and operators,such as the unreasonable volume of regional bicycles,the imbalance of supply and demand of vehicles in various regions during peak hours,and the im balance of operation and management.The key to solving these problems lies in the analysis and prediction of the future demand for bicycles in the region,which is of great significance for the vehicle scheduling of shared bicycles and the operation and management of operators.This paper firstly visualizes the shared bicycle data,including the distribution of rentals,the impact of time characteristics(such as seasons,months,hours,etc.)on the rental volume,and the impact of environmental characteristics(such as temperature,humidity,wind speed,etc.)on the rental volume.Then,in the feature engineering part,the vacancy outliers of the data are filled,and the feature analysis is performed by the visual analysis result.The influence of time fa ctor on the amount of car rental is obvious,so the time characteristics of the car rental,the environmental time characteristics and the periodic characteristics are constructed based on the time attribute.In addition,a series of features are construct ed for the characteristics of the morning and evening peak periods of the day.And make feature selection.Finally,Support Vector Regression model,Random Forest regression model and GBDT regression model are established to predict the demand of shared bicycles.And using the stacking method to fuse the three models,propose to establish SSRG prediction model,and compare and analyze their respective prediction effects.The experimental results show that the combined SSRG model has better data fitting de gree than the traditional single model and the prediction accuracy is higher.