<p>Commuting travel behavior analysis can provide support for policy formulation, transportation planning, and bus route optimization. Existing works on commuting travel behavior analysis focus on commuting travel pattern recognition, while there is a lack of discussion on the heterogeneity of commuting travel behavior itself. In this study, we identify the entire sample of commuting vehicles in the road network based on license plate recognition (LPR) data, and then propose a new travel regularity analysis framework by using self-supervised representation learning with contrastive learning. The framework proposed in this study can directly learn travel semantics from trajectory data without the need for label information. The analysis results based on LPR data indicate that the framework proposed in this study can effectively measure travel regularity. Based on the random forest and SHapley Additive exPlanations methods to analyze the influencing factors of commuting vehicle travel regularity, it is found that there is a correlation between the travel behavior characteristics and travel regularity. Compared to travelers with weak or medium travel regularity, vehicles with strong travel regularity are more likely to exhibit the following travel characteristics: a moderate detection frequency in the morning, a low detection frequency in the afternoon, a moderate full day outdoor time, and the first detection occurring in the morning.</p>