Combined Long-Term and Short-Term Predictions of Cut-in Intention towards Vehicle Platoons

Vehicle platoons are groups of vehicles traveling together in close formation. In mixed traffic flow, they can frequently encounter cut-ins of human-driven vehicles (HDVs). In this paper, to predict the driver cut-in intention towards the vehicle platoon, we propose a combined prediction system by integrating a long- and a short-term prediction algorithm. The two algorithms, both constructed based on long short-term memory (LSTM) networks, are developed to identify the driving characteristics involved in the cut-in preparation and execution phases, respectively. Cut-in experiments in the platooning context are conducted to train and test the proposed method. Offline classification tests are used to show the advantage of LSTM over two other machine learning methods for both long- and short-term predictions of the cut-ins. The pseudo-online testing method is designed to evaluate the prediction performance of the proposed method. The results suggest that the combined prediction system benefits from the strengths of both the long- and short-term prediction algorithms, which can achieve a prediction accuracy of 95.8% with average and maximum prediction horizons being 3.6 and 11.4 s, respectively. Finally, we illustrate the prediction process of the proposed system in online testing experiments to show its effectiveness in the real-time prediction of the driver cut-in towards the vehicle platoon.