Examining the sequential patterns of self-regulated learning (SRL) behaviors is gaining popularity to understand students' performance differences. However, few studies have looked at the transition probabilities among different SRL behaviors. Moreover, there is a lack of research investigating the temporal structures of students' SRL behaviors (e.g., repetitiveness and predictability) and how they related to students' performance. In this study, 75 students from a top North American university were tasked to diagnose a virtual patient in an intelligent tutoring system. We used recurrence quantification analysis and sequential analysis to analyze the temporal structures and sequential patterns of students' SRL behaviors. We compared the differences between low and high performers. We found that low performers had more single, isolated recurrent behaviors in problem-solving, whereas the recurrent behaviors of high performers were more likely to be part of a behavioral sequence. High performers also demonstrated a higher transition probability across the three phases of SRL than low performers. In addition, high performers were unique in that their behavioral state transitions were cyclically sustained. This study provided researchers with theoretical insights regarding the cyclical nature of SRL. This study has also methodological contributions to the analysis of the temporal structures of SRL behaviors.