Study On Prediction Model Of Shared Bicycle Demand Based On Machine Learning

Bicycles have always been an important tool in urban transportation.With the booming of the mobile internet era,shared bicycles have become popular in urban transportation.It not only facilitates people's travel,but also promotes the development of sharing economy.However,the unreasonable investment on the shared bicycle resources has caused a great impact on shared bicycle company's economic interests and the city cityscape.This paper designs a combined forecasting model of shared bicycle demand through analyzing the data of shared bicycle usage and guide shared bicycle reasonable placement by predicting the demand of shared bicycles.The main tasks include:First preprocess data and visually analyze data to obtain the influencing factors affecting the usage of shared bicycles.On this basis,screen the features by the backward removal method,and determined the data of the final construction model by combining the results of the visual analysis.Then study on traditional prediction algorithms,and focus on Gradient Boosting Decision Tree(GBDT)algorithm,random forest algorithm and BP neural network algorithm.And using the experimental data determined previously to implement the single prediction model.The experimental results show that The same feature is different in the GBDT model and the random forest model,and the temperature has the greatest impact on the demand for shared bicycles.The R-square values of the three single-term prediction models are similar,and the prediction accuracy has a lot of room for improvement.On this basis,combine three single prediction models to design two combined prediction models,one is a linear combination prediction model with the smallest sum of prediction errors,the other is the nonlinear combination model based on BP neural network.Experiments on two combined models show that the accuracy of the two combined forecasting models is improved compared with the single forecasting model,but the nonlinear combined forecasting model has better prediction effect,mainly because the linear relationship in the linear combination limits the degree of optimization of the combined model to some extent,the relationship between the single prediction models in the combined model has certain complexity,not just a linear relationship,but the BP neural network can mine its nonlinear information while satisfying the linear relationship to make model is more effective,and its R-square value is increased by 0.02 compared to the linear combination prediction model.Therefore,the nonlinear combination model based on BP neural network is used as the final shared bicycle demand forecasting model.