Research On Taxi Demands Forecasting Of Urban Regions

Taxi has always been one of the most important public traffic services in modern society.Due to the impact of the advancement of the Internet and modernization,people become more willing to use online car Hailing for Taxi services.However,a lot of problems come with the rising popularity of online car Hailing such as extended waiting time for both the passengers and the drivers,causing serious issues like wasting resources,and terrible experience for the passengers.If the demand for taxis can be accurately predicted,the online taxi Hailing platform can dispatch or guide taxis in advance,which can greatly reduce the waste of resources,and provides not only convenience for the passengers but also extra revenue for the drivers.Taxi demand prediction in urban areas has become a hot topic in Intelligent Transportation.There is a spatula correlation between the taxi demand in urban areas and regions from a spatial dimension perspective.In addition,the dynamic distribution of people and vehicles in urban areas and the continuous movement in the spatial dimension also reflect the complex spatial dynamic correlation.Human behavior is inherently timedependent from the time dimension perspective.If both time and space dimensions are considered,there will be a temporal-spatial correlation between the movement of people and vehicles between different areas of the city over time.Therefore,We propose a method using the deep neural network to learn the spatial-temporal dependence and predict the taxi demand in urban areas based on the historical taxi order data.The work of this paper mainly includes the following aspects:(1)Based on taxi order data in Manhattan,New York,we built regional spatial correlation based on taxi history data,building spatial correlation,convenience,and similarity among urban areas in the spatial dimension,and achieving spatial correlation modeling.(2)A forecasting model for taxi demand in urban areas is proposed,which combines the context-gated loop neural network with the multigraph convolution neural network,calculates the weights of observations at each time step using the attention mechanism,captures the global information in the city using the context-gated loop neural network with shared parameters,and finally uses several convolutions to learn and fuse the proposed regional spatial correlation.Thus,a regional taxi demand forecasting model is constructed,which can fully mine the space-time information.(3)A visualization system of taxi demand prediction in urban areas is designed and implemented.The POI information in Manhattan is obtained through the visualization system,and the taxi demand and forecasting results are displayed visually.The system design uses front-end and back-end separate architecture,the back-end uses Spring Boot framework,the front-end uses VUE framework,and calls JS interface of Baidu Map and Chart to visualize the historical demand and forecast results of the whole urban area and each region,which can be used for the observation analysis and prediction of taxi demand in urban areas.