Development of Moped-Following Models by Characterizing the Riding Style of Moped Riders

In recent years, mopeds have emerged as attractive two-wheeled vehicles because of their high mobility. However, the flexibility and randomness associated with mopeds make the development of two-wheeled microscopic simulation models more difficult. Human factors (HFs), for example, the riding style, are the major source of behavior heterogeneity, and also a reason for flexibility. As such, this paper proposed two riding style recognition methods: clustering analysis; and the topic model. Clustering analysis identified the riding styles according to the aggregated behavioral parameters of each rider, whereas the latent Dirichlet allocation (LDA) model, a classic topic model, was based on time step level. The LDA model can recognize riding styles according to riding behaviors at each time step in the whole moped-following period, considering different following conditions. The riding styles categorized by the two methods were incorporated into a mainstream moped-following model. The model error results demonstrated the necessity of taking HFs into account. The error of the moped-following model considering the riding styles obtained by the LDA model was lower than that obtained through clustering analysis by 6.58%. Thus, it was demonstrated that the LDA model can provide a more precise understanding of the riding styles because these were more reasonably identified. In addition, the most critical behavioral characteristic parameter that distinguishes the riding styles was found, that is, speed. This study provides deep insights to assist a more precise understanding of riding styles and supports the development of a moped-following model that reproduces the behavioral characteristics of moped riders better.