Research On Traffic Forecast And Allocation Of Shared Bicycles

With the sharing economy rapid development and the low-carbon environmental protection concept becoming more and more popular,shared bicycles have been favored by urban residents with their efficiency,convenience and energy saving.However,due to the groupness and residents’ transportation periodicity,there is a tidal phenomenon leading to an imbalance between the bicycle supply and demand in the urban areas with different functions and attributes,which puts certain pressure on public transportation.Therefore,predicting the traffic flow in a certain area,formulating a reasonable and scientific bike-sharing allocation plan,and rebalancing the supply and demand in the bicycle system can impove the residents’ experience,increase the bicycle utilization rate,optimize the transportation resources distribution and promote the sustainable urban transportation development.The most widely used allocation method is the optimal path allocation method,which has high time complexity and is sensitive to the bicycle system scale.For multiple traffic conditions in different cities,adapting to local conditions is an important prerequisite for the successful deployment strategie implementation,which requires accurate future bicycle traffic forecast.Therefore,this paper focuses on bicycle demand forecasting and allocation scheme generation,uses deep learning algorithm,and provides the allocation scheme generation method,which is driven by bicycle leasing data,and dynamically deploys bicycles to meet bicycle demand.The specific research content includes the following three aspects:(1)Bicycle delivery area generation based on unsupervised learning.Based on the bicycle leasing data,this paper excavates the tidal bicycle traffic phenomenon dependence on the spatiotemporal bicycles distribution.Based on the bicycle location coordinates,this paper excavates the spatial bicycle location distribution relationship by clustering method,automatically generates bicycle delivery areas,analyzes the demand and deployment characteristics of shared bicycles,and generates bicycle inflow sequences and outflow sequences in each delivery area,which are used for bicycle demand forecasting and dynamic allocation research.(2)Bicycle flow sequence forecast based on similarity-based embedding combination Informer(SEC-Informer).The bicycle demand forecast problem is transformed into the inflow and outflow sequence predicting problem in each delivery area.The relative position code,global position code and cycle position code are combined with the bicycle flow information to synthesize the sequence periodicity information.This paper proposes SEC-Informer,which is the improved model of Informer.SEC-Informer replaces the addition-based embedding combination method with the similarity-based embedding combination method,which adjusts the attention score based on the similarity between location codes and improves the model forecast accuracy.Through comparative experiments and ablation experiments,SEC-Informer has higher forecasting accuracy than Informer,which can effectively predict bicycle demand.(3)Allocation scheme generation based on nomad algorithm with constraints(NCA).Model the cycling allocaion problem as an optimization problem constrained by the bicycles flow and aiming to minimize operational losses.Compared with the unconstrained nomadic algorithm,NCA improves the local search and global optimization strategies,and optimizes the tribal initial positioning method.Based on the predicted stock and transfer volume,the interregional bike-sharing allocation scheme in different time periods is obtained.The experimental results of the shared bicycle dataset in Shanghai and New York show that the nomadic algorithm can obtain a better bike-sharing allocation scheme than the traditional allocation algorithm.