The temporal invariants reflect the temporal logic relationship between events and have been widely used in anomaly detection, system behavior understanding, model reasoning, and other techniques. Generally, mining temporal invariants through analyzing the log data of software system is in actual use. Compared with totally ordered log, partially ordered log can provide a more accurate data source for mining algorithm. However, the existing temporal invariants mining methods based on partially ordered log have some problems such as its low efficiency. For this reason, this study uses the system execution path as the data source and proposes a temporal invariants mining method SOTIMiner based on set operations and studies an improved scheme. Compared with existing methods, it does not need to traverse the log data in reverse and for the reason that it has a higher efficiency. Experiments show that the method's average efficiency is 3.23 times of the Synoptic mining tool on the basis of guaranteeing the same result.