Short-term Travel Demand Forecast Of Free-floating Shared Bicycle Based On Stacking Strategy And Dispatching Route Optimization

Under the background of the country’s promotion of low-carbon transportation and green travel,more urban residents choose to use shared bicycles as a slow-moving vehicle to solve the last mile problem.During the new coronavirus epidemic,shared bicycles have become an important travel mode in the urban transportation system that can undertake a complete travel chain due to their convenience and safety.However,with the rapid development of shared bicycles,the temporal and spatial imbalance of residents’ travel demand has led to problems such as imbalance between bicycle supply and demand,low turnover efficiency,and many broken bicycles.Therefore,reasonable deployment of bicycles on the basis of accurately predicting the travel demand of shared bicycles is of great significance for improving the operating efficiency of the shared bicycle system and customer satisfaction.This article mainly studies how to improve the accuracy of the free-floating shared bicycle demand forecasting and the efficiency of bicycle deployment through ensemble learning algorithms and scheduling route optimization models.We use a data-driven approach to analyze the temporal and spatial characteristics of shared bicycle travel demand.Then we explore the impact of external factors such as weather on the fluctuation of bicycle demand,and establish a combined bicycle shared travel demand forecast model based on the Stacking strategy.Considering the predicted value of dispatching demand and the recycling of broken bicycles,and paying attention to dispatching costs,a model of route optimization for shared bicycle dispatching is constructed.The main research contents are as follows:(1)Based on more than 3.21 million cycling data of Mobike in Beijing for 21 days,this paper studies the temporal and spatial distribution characteristics of shared bicycle travel demand.On the basis of elaborating the structure of the shared bicycle system and data preprocessing,we analyzed the impact of time-of-day and day-of-week on the demand for shared bicycle travel.Through projection and regional meshing methods,we analyzed the differences in the demand for shared bicycles in different areas of land properties.(2)Through causal analysis and grey relational analysis,we realized the feature selection and sample set construction of the prediction model,thus laying the foundation for the construction of the shared bicycle travel demand prediction model based on integrated learning.First,we used the Granger causality analysis method to analyze the causal relationship between weather variables and bicycle demand,and screened out weather influencing factors that are useful for predicting the demand for shared bicycle.Considering the similarity between the weather features,this paper uses the gray correlation index to filter out the sample sets of similar days with high similarity.The results of the case analysis show that the method of selecting similar days through gray correlation can effectively reduce the prediction error of the primary learner.For example,the mean square error of the random forest model using the similar day selection method is 33.1% lower than that of the traditional adjacent day method.(3)A combined forecasting model of shared bicycle demand based on Stacking strategy is proposed.Based on the selected similar day training sample set,this paper constructs an uncertainty integration model based on neural networks,random forest regression and multiple linear regression algorithms.Taking the cycling data of Mobike users in Beijing for example analysis,we analyze the influence of the number of primary learners and different fusion strategies on the prediction accuracy of the ensemble model,and determined the optimal integration framework.The results show that compared with the neural network model with the highest prediction accuracy in the primary learner,the average absolute percentage error of the combined forecasting model of bicycle demand based on the optimal integration strategy is reduced by 9.1%.The integration strategy improves the prediction accuracy and generalization performance of individual learners.(4)Based on the predicted value of shared bicycle travel demand and considering the identification of broken bicycles,this paper establishes a shared bicycle scheduling route optimization model.First of all,based on the analysis of the reasons and methods of shared bicycle scheduling and other related theories,we realized the effective identification of troubled bicycles by establishing a Cox proportional failure model.Then,considering the comprehensive dispatch cost,we established a bicycle dispatching route optimization model for good bicycles and broken bicycles.The results of the case analysis show that the route optimization model can reduce the dispatch cost and provide a reference for the actual bicycle dispatch process.