An innovative method to measure and predict drivers' behaviour in highway extra-long tunnels using time-series modelling

Previous studies lack a comprehensive evaluation model that combined the subjective perception of the driver and the objective driving environment. This work investigates the characteristics of drivers' behavior risk in highway extra-long tunnels. Real-vehicle tests were conducted in two typical extra-long tunnels and the speed of skilled and unskilled drivers were collected simultaneously. The quantified model of drivers' behavior risk was proposed based on the safety speed difference. The variation characteristics of behavior risk both inside the tunnel and ordinary highway were analysed. Further, the NARX neural network was used to predict real-time speed with the heart rate regarded as the input variable. Results showed that skilled drivers showed the highest behavior risk in the internal zone, while the highest value of unskilled drivers was at the exit zone in the tunnel section. Both two types of drivers presented the highest and the lowest behavior risk on the ordinary highway and the tunnel entrance zone respectively. The proposed NARX model could predict synchronous speed with high accuracy. These results of the present study concern the driver's risk characteristics in Internet ofVehicles and howto establish the automated driver model in the simulation driving environment.