Research On Short-term Demand Forecasting Of Shared Bicycles Based On Site

The emergence of bicycle sharing has maximized the accessibility of public roads.It not only eases the pressure of urban traffic and satisfies people’s demand for short-distance travel,but also solves the problem of the last mile from commuters’ residence to the subway.With the rapid popularity of bicycle sharing,the problem of unbalanced number of bicycles in the stations caused by its unique spatial and temporal demand fluctuation is worsening.On the one hand,the bicycles in some stations are idle due to "oversupply" and even have nowhere to be placed,which not only affects the cityscape,but also wastes resources and increases the operating costs of shared bicycles.On the other hand,the shortage of bicycles in some stations not only affects the user’s experience,but also reduces the utilization rate of bicycles,which again increases the operating cost of shared bicycles.Therefore,achieving smooth,orderly,healthy,green and lasting development of shared bicycles has been a hot issue for research in the recent years,and has important practical significance for promoting the development of the sharing economy and environmental protection.Accurate short-term demand prediction of shared bicycles within a site is the key to solve the unbalance distribution of shared bicycles among sites.In this thesis,we aim to accurately predict the short-time demand of regional shared bicycles and study an effective and feasible method to predict the short-time demand of regional shared bicycles based on the prediction algorithm-Light GBM algorithm.The main contributions of this thesis are listed as follows:(1)Analyzing and extracting the main characteristic factors of affect to shared bicycles short-term demand.Based on the cycling data in Bike Sharing Data Set,a dataset in the UCI machine learning repository,firstly,the temporal and spatial feature factors associated with it are studied by processing the cycling records and mining the demand patterns of bicycle use at stations.Among them,the temporal characteristics are obtained by studying the riding patterns of bike rental users in different time periods;the spatial characteristics are obtained by studying the riding patterns of bike rental users in different environmental conditions.Secondly,the Spearman correlation coefficient is used to calculate and analyze the degree of association between different feature factors,and then extract the feature factors that have significant influence on the short-time demand of shared bicycles.(2)Proposing a model of Light GBM based on Bayesian optimization to predict shared bicycle short-term demand.Firstly,we systematically study the series of integrated learning regression prediction models RF,GBDT and XGBoost and focus on their improvement process.Based on this,we construct a Bayesian optimization-based Light GBM based on the prediction algorithm-Light GBM for the influence of bike-sharing timing and spatial characteristics on regional bicycle distribution by using Bayesian algorithm to optimize the Light GBM hyperparameters.Light GBM short-time demand prediction model for shared bicycles.Second,the performance of the constructed prediction model is evaluated by comparing and analyzing with random forest and XGBoost machine learning models using root mean square error(RMSE),mean of absolute error(MAE)and squared correlation coefficient(R2).The results show that the Bayesian-optimized Light GBM model has higher prediction accuracy and stronger model generalization ability for the problem of uneven distribution of bike-sharing areas.(3)Designing and implementing a shared bicycles shore-term demand prediction system with Bayesian optimization-based Light GBM to use.The Light GBM Bayesian optimization-based bicycle sharing short-time demand prediction algorithm is combined with concrete practice to design and implement a bicycle sharing management system for guiding bicycle sharing resource reallocation practice and improving the effectiveness and practicality of bicycle sharing services.