Smart travel time prediction model for urban traffic using long short-term memory network

Abstract

Traffic jams or traffic congestion is the most important factor of the urban road networks for advanced travel plans and managing the traffic flow. It causes service holders, travellers, riders, drivers and logistics to not reach their desired location in a timely manner. Traffic congestion cause’s real-time traffic information to become a crucial part of advanced traveller information systems (ATIS) for daily travelling and smartly managing vehicle flow. An intelligent transportation system (ITS) is more essential for urban road networks. Travel time and traffic speed are major components in ITS to assist the travel planning and proactively managing traffic in the urban road networks. The study investigated the Google Maps traffic information in various aspects like Google Maps traffic information accuracy, multi-steps ahead travel time and traffic flow prediction on various urban road networks. Real-time traffic data were collected from Google Maps for several urban road networks of Dhaka city. A long short-term memory (LSTM) network model is developed to perform the multi-step ahead travel time and traffic flow prediction based on the historical traffic and weather data. The experimental results explored that, Google Maps provided current traffic information accuracy average 91.19%. The proposed LSTM model's mean relative error is varying between 5.84% ~ 10.93% for one-hour advance travel time prediction. Models predicted traffic flow error range 8.25% ~14.09% on three different road sections after mapping predicted traffic speed by using time-dependent correlation (TDC). Also, results showed that the proposed model prediction accuracy improves and is stable with the smaller time interval. The proposed prediction model is reliable for predicting multi-steps ahead of travel time and traffic flow prediction. It will be used in transportation systems road network planning, design, traffic policy-making, proactively managing traffic flow and smartly assist traffic signals management. Moreover, the proposed models are essential for smart travel planning, finding congestion and best route, time-saving, avoid traffic jams, cost-minimizing and ensure more trips for riders, drivers and logistic services in urban road networks.