A risky prediction model of driving behaviors: especially for cognitive distracted driving behaviors

Abstract

The non-driving related operation behavior in driving process has a significant impact on road traffic status and driving safety, but there is less systematic study on the main characteristics and influence mechanism of such behaviors. Aiming at this problem, four types of typical behaviors of normal and abnormal driving are monitored and recorded by real vehicle test. The cognitive distracted driving behavior is taken as the research object, and the influence mechanism and prediction method of distracted driving are studied by using the driver's physiological state and vehicle running state. This paper focuses on the changes and statistical characteristics of driver's physiological state parameters and vehicle running state parameters during distracted driving, and then explores the influence mechanism of different types of distracted driving tasks with different loads on driver's state. This paper analyzes the influence mechanism from two aspects of human and vehicle. Based on the comparison of behavior criterion and load criterion, the parameter system of cognitive distracted driving behavior considering driving load is obtained after cross analysis. The prediction model is established as the training sample of LSTM model, and the model is tested with the data collected from real vehicle test After 100000 iterations, the training accuracy is 90.2% on the training set and 74% on the test set. The results showed that the cross-comparison method is scientific and reasonable, and the prediction model of distracted driving behavior based on physiological state and vehicle running state has good accuracy.