Driving style recognition considering speeding behavior under different working conditions

Abstract

The driving style of the driver has a significant impact on the safety of vehicle operation. This paper proposes a driving style recognition model that takes into account speeding behavior, aiming to improve the accuracy of driving style recognition. Initially, vehicle operation data is collected through on-road experiments with drivers. Subsequently, feature parameters related to driving conditions are extracted from the vehicle operation data, and dimensionality reduction is applied to these parameters. The principal components extracted are then utilized as inputs for the particle swarm optimization support vector machine algorithm to determine driving conditions. This information is used to establish the speeding threshold, which is then used to calculate the number of speeding occurrences and the longest speeding time as evaluation indicators. These indicators are integrated into a comprehensive evaluation system comprising 18 evaluation criteria to improve the accuracy of driving style recognition. Lastly, the particle swarm optimization support vector machine algorithm and convolutional neural network algorithm are employed for driving style recognition. The results indicate that the particle swarm optimization support vector machine algorithm demonstrates fewer iterations and higher accuracy, reaching 97.4%. Furthermore, both algorithms show improved accuracy in driving style recognition when considering speeding behavior, affirming that the inclusion of speeding behavior enhances the accuracy of driving style recognition.