Spatiotemporal Traffic Flow Forecasting for PHETs Based on Data Mining and Deep Learning

摘要

Plug-in electric vehicles (PEVs) are promoted as environmental-friendly future vehicles. Comparing to the conventional vehicles, they refuse emissions of carbon dioxide and harmful gases while consume no fossil fuel. This paper presents a spatiotemporal flow 3D model and a spatiotemporal filter method to clean the raw data and adapt several types of convolutional neural networks (CNNs) to predict the traffic flow of plug-in hybrid electric taxis (PHETs). The 3D model divides the area into cells and directly show the traffic flow levels in each minute and each cell. By adapting the filter, the cleaned data clearly indicates the main trend of traffic flow and have no obvious outliers. Besides, based on almost 30 million orders of over 30 thousand taxis and with the help of the recent neural networks such as VGG and Resnet, our prediction is very close to the reality. Although the research only considers the traffic flow prediction in Beijing, the methodological framework is adaptive to other cities with similar dataset available and provides reference for dispatching charging opportunities to PHET and planning time-saving routes to vehicles.