Lipases with high thermal stability have greater application in industrial production. Focusing on wild-type lipase and their thermotolerant mutants, this study employs dynamic graph embedding to explore the change of spatiotemporal characteristics in residue-residue interactions, analyzing the relationship between residue mutations and the thermal stability of lipase. K-means clustering is used to analyze the synergistic relationships between residues, revealing that thermotolerant mutants achieve higher thermal stability by enhancing the synergistic interactions between secondary structures, resulting in a more balanced and tightly-knit internal structure. Critical residues are identified using anomaly node detection methods, showing that residues with spatiotemporal instability are mainly distributed on highly flexible loop regions. Specifically, mutations at Tyr89 and Asp132 restrict the motion range of the loops, making the lipase structure more stable. The study demonstrates that mutating highly flexible regions to increase structural rigidity can effectively enhance the thermal stability of lipases.