In　the　transient　stability　assessment　(TSA)　problem　of　a　power　system,　the　limited　data　mining　ability　of　common　machine　learning　algorithms　prevents　further　improvement　of　TSA　model　evaluation　accuracy.　To　solve　this　problem,　taking　different　instability　patterns　of　power　system　data　as　the　entry　point,　a　TSA　method　based　on　adaptive　capture　of　instability　patterns　is　proposed.　An　adaptive　stability　judgment　combination　model　is　constructed,　one　which　is　composed　of　multiple　sub-evaluation　models　and　an　instability　pattern　discrimination　model.　First,　the　original　data　set　is　classified　according　to　the　different　instability　patterns,　and　several　sub-evaluation　models　are　trained　for　different　instability　patterns.　Then,　the　weight　value　of　the　instability　pattern　discrimination　model　output　is　used　to　integrate　the　sub-evaluation　model,　and　the　instability　pattern　of　the　input　data　is　captured　adaptively.　Finally,　a　case　study　is　performed　on　the　IEEE39-bus　system　and　the　East　China　power　grid　system.　Simulation　results　show　that　the　proposed　method　can　reduce　the　negative　rate　of　unstable　samples　and　further　improve　model　evaluation　accuracy,　which　verifies　the　effectiveness　of　the　proposed　method.　©　2024　Power　System　Protection　and　Control　Press.　All　rights　reserved.