The emergence of taxi requesting service have changed the situation of traditional taxis. Nowadays, it has gained great popularity all over the world. As more people use this service, some problems spring up gradually and the imbalance between supply and demand is one of the most serious ones. It is urgent for researchers to solve this problem because it affects greatly the service quality of taxi system. In summary, traditional papers involving the demand prediction of taxi-hailing are mainly divided into two directions. Some researchers tend to use regional attributes (e.g., land use variables) for spatial modeling and analysis (e.g., using geographically weighted regression, GWR) and then interpret the model. The others mainly depend on spatiotemporal correlation to predict the demand (e.g., autoregressive integrated moving average, ARIMA) and then focus on the improvement of prediction. Based on previous researches, this paper proposes an adaptive multi-view deep learning model which comprehensively integrates the focuses of previous studies. The framework of the model includes four views, namely feature views (including various features of zones such as the conditions of weather, transportation, land use and so on), semantic views (division of zones that functionally similar), spatial views (searching for adaptive spatial neighbors), temporal views (searching for adaptive time windows). By embedding these views into our model, the accuracy and extension of prediction are greatly improved. To evaluate the prediction performance of our model, it is respectively compared with models which use different algorithms or have different model frameworks. Validation based on large amounts of data shows the superiority of our proposed model.
Prediction for Taxi-Hailing Demand—An Adaptive Multi-view Deep Learning Model
Lect. Notes Electrical Eng.
International Conference on Green Intelligent Transportation System and Safety ; 2021 November 19, 2021 - November 21, 2021
2022-10-28
17 pages
Article/Chapter (Book)
Electronic Resource
English