An improved Markov model for idle time prediction based on spatiotemporal learning

In order to reduce the fuel consumption of idling, lots of vehicles are equipped with idle start-stop system. In the actual traffic situation, heavy traffic result in frequent short time idle stop, which causes two major problems with the existing start-stop system, frequent starting and stopping and invalid stopping. That not only exacerbates component wear but also makes fuel efficiency worse. Aiming at this problem, an improved Markov model for idle time prediction based on spatiotemporal learning is proposed. Based on original Markov model, related training subset for training model is determined according time and space information, which could increase prediction accuracy by reducing the redundant noisy data. Experiments conducted on real urban dataset demonstrate that our proposed model could get better performance, the F1-measure increased by 8.0%. Start-stop strategy using the prediction method can make invalid idling reduced by 35.1%, which can effectively avoid invalid idling and obtain a reduction of frequent starting and stopping.