New method for predicting long-term travel time of commercial vehicles to improve policy-making processes

Abstract Long-term travel time prediction, ahead of making a trip, is vital from the planning perspective of delivery freight, timetable design, vehicle/crew scheduling and further activities. The better the prediction is, the higher the reliability of service that can be offered. This study presents a discrete and continuous combined analysis for attaining improved long-term travel time prediction (LTTP) of commercial vehicles. One main problem of LTTP is that the speed factors are unknown ahead of trips. In light of this, the nonnegative tensor factorization and completion with neural weighted initialization is proposed to extract the potential speed patterns among multiple discrete factors and to complete the sparse tensors. The Gaussian mixture regression is adopted for handling the continuous factors. The proposed methodology with a combined discrete and continuous analysis is able to effectively integrate multiple factors into the computation, including vehicle type, road type, days, time period, weather conditions, driver differences and travel distance. The methodology is able to reduce the long-term travel time prediction error between 14% and 43% compared with the traditional average speed method and other baseline methods, which suggests its effectiveness. It can strategically assist policy-making processes of stakeholders on investment, insurance, planning and management, and can help tactically in predicting long-term travel time ahead of the scheduled trips to improve the reliability of the schedules. Furthermore, operationally, it can also be used to enrich current navigation information systems by separately predicting the commercial vehicles’ travel time based on multiple factors.