Demand Prediction And Scheduling Method Of Shared Bicycle Based On Deep Learning

With the advancement of the concept of carbon neutrality and emission peak,green,environmentally friendly and low-carbon slow-moving transportation systems are loved by people deeply.In the process of building the urban slow traffic system,the proportion of bicycle trips is also gradually increasing.As a product of the sharing economy and Internet mobile payment,shared bicycle has developed rapidly because of their convenience and speed.To a large extent,shared bicycle meets the public’s demand for short-distance travel,and also leads the public to form a green travel and low-carbon life.However,many problems have also been exposed in the process of sharing bicycles in the market operation,such as a large amount of waste of public resources,random parking and occupying other public resources,affecting traffic order,and uneven distribution of bicycles in time and space,which is difficult to meet user needs.These issues are also new challenges for urban development.Therefore,the deep learning methods was used to predict travel demand and optimize scheduling accurately,and provide a basis for the refined management of shared bicycle operators.The research work and conclusions of each part are as follows.（1）Preprocess and analyze the acquired historical riding data of Citi Bike shared bicycles,and study the influencing factors and spatial distribution characteristics of shared bicycles.According to the historical riding order data,the age,gender,riding duration and other factors that affect the user’s travel flow are analyzed in different time dimensions（working days,non-working days,24 hours）,and the distribution of passenger flow in time is obtained.The occurrence and attraction of different parking sites are obtained by sub-aggregating the order data.（2）Use the time series decomposition method to decompose the original shared bicycle traffic sequence into trend sequence,seasonal sequence and residual sequence,and bring each sub-series into the long-short-term memory neural network model for prediction.The resulting forecasts are merged into the final forecast.The long-shortterm memory neural network and gated recurrent unit without decomposition are selected to compare with the prediction method proposed in this paper.The comparison results show that the prediction result of the model proposed in this paper is closer to the real value,and the prediction accuracy is higher.（3）A scheduling optimization model of shared bicycle is established to minimize the scheduling cost.In order to adapt the ant colony algorithm to the shared bicycle scheduling problem,the actual scheduling amount is introduced to improve the ant colony algorithm to solve the scheduling optimization model,and the area with large demand for shared bicycles is selected to verify the feasibility of the algorithm.