Automotive Science and Engineering

Research on vehicle longitudinal control with a stop and go system is presently one of the most important topics in the field of intelligent transportation systems. The purpose of stop and go systems is to assist drivers for repeatedly accelerate and stop their vehicles in traffic jams. This system can improve the driving comfort, safety and reduce the danger of collisions and fuel consumption. Although there have been many attempts to model stop and go maneuver via traffic models, but predicting the future vehicle's acceleration in steps ahead has not been studied much in this models. The main contribution of this paper is in designing integrated genetic algorithm-artificial neural network (GA-ANN) which is a soft computing method to simulate and predict the future acceleration of the stop and go maneuver for different steps ahead based on US federal highway administration’s NGSIM dataset in real traffic flow. The results of this study are compared with two methods, back propagation based artificial neural network model (BP-ANN) and standard time series forecasting approach called ARX model. The mean absolute percentage error (MAPE), root mean square error (RMSE) and coefficient of determination or R-squared (R2) are utilized as three criteria for evaluating predictions accuracy. The results showed the effectiveness of the proposed approach for prediction of driving acceleration time series. The proposed model can be employed in intelligent transportation systems (ITS), collision prevention systems (CPS) and driver assistant systems (DAS) such as adaptive cruise control (ACC) and etc. The outcomes of this study can be used for the automotive industries who have been seeking accurate and inexpensive tools capable of predicting vehicle speeds up to a given point ahead of time, known as prediction horizon, which can be used for designing efficient predictive controllers based on human behaviors.