Demand Forecast And Self-Balancing Trip Planning Of Public Bike System

The Public Bike System(PBS)is an important part of urban transportation.The growing demand of users brings great challenges to PBS maintenance personnel.It is of practical significance to study the imbalance between supply and demand of PBS for urban governance and improvement of urban service quality.At present,public bicycle demand prediction mostly uses traditional neural network methods,which is difficult to capture the spatial-temporal correlation in the data,and the prediction results are difficult to meet the accuracy requirements.The public bike trip planning can be used to navigate users in mobile computing,but its algorithm research is less.And it is not considered enough to guide the flow of people and improve the continuous service ability of the system.In view of the above problems,this thesis analyzes and studies on public bicycle demand forecasting and self-balancing travel planning,and proposes two methods.(1)A Multi-Period Spatio-Temporal Multi-Graph Convolutional Networks prediction model(MP-STMGCN)is proposed.The model contains multiple spatiotemporal convolution modules and two fully connected layers.The spatiotemporal convolution module includes three layers of neural networks.The first layer is the attention mechanism layer,which is used to enhance the module’s ability to capture data features.The second layer is a multi-graph convolution layer,which constructs multiple graphs to encode various non-Euclidean correlations and applies it to multi-graph convolution.The third layer is the time-dimensional convolution layer,which used to capture spatial-temporal correlation.Considering that the site requirements are related to the time period,this thesis divides the data into multiple periods(recent,daily-periodic,weekly-periodic)as model input.Applying multiple modules and one layer of full connection layer,three predicted values are obtained.Finally,the final predicted values are calculated by weighted sum of these three predicted values through the last layer of full connection layer.In this paper,prediction experiments are performed on realistic bike flow datasets,Hangzhou public bike data set(HZ)and New York public bike data set(YNC),and compared with the existing seven prediction methods.The results show that our MP-STMGCN model is superior to other methods.On both data sets,compared with the best ASTGCN prediction results in the benchmark method,the mean absolute error(MAE)performance reduced by 7.56%and 8.53%respectively,and the mean square error(RMSE)indicators reduced by 11.76%and 6.97%.(2)A priority based self-balancing travel planning strategy(Priority awarded Trip Planning,PTP)is proposed,which aims to guide the users to choose the renting and returning stations,to achieve the purpose of meet the users’ demand.The strategy,firstly,calculates users’ priority and stations’ priority,and then matches users and stations according to priority to achieve public bike trip planning,thereby balancing station resources and improving PBS’s continuous service capability.A large number of simulation experiment results show that:PTP strategy saves 25.37%of the average trip time-cost compared with the benchmark algorithm OTP(Original Trip Planning);at 90k user scale,PTP strategy helps 95.31%of users successfully use the PBS service,which is better than the traditional GTP(Greedy Trip Planning)algorithm in successful trip planning ratio.And the algorithm complexity of PTP is one order of magnitude lower than GTP.In summary,this paper,firstly,analyzes the spatial-temporal correlation of station demand based on Hangzhou public bike data set;proposes an MP-STMGCN prediction model for predicting the station-level demands.And then proposes a priority-awarded public bike trip planning strategy PTP to achieves system-wide self-balancing.Afterwards,the results on realistic bike flow datasets show that both of our methods are superior to other methods.Finally,the paper looks at the three factors that demand factors,the threshold conditions of public bike trip planning and dynamic scheduling.