By　training　the　running　data　output　in　the　data　simulation　system,　an　abnormal　condition　monitoring　system　model　based　on　three　data-driven　algorithms　such　as　support　vector　machine,　local　outlier　factor,　and　isolated　forest　algorithm　is　designed,　and　the　training　results　are　evaluated.　A　set　of　accident　condition　data　is　selected　to　test　the　trained　model　and　evaluate　the　test　results.　The　results　show　that　the　three　training　results　are　basically　consistent　with　the　actual　situation,　and　the　accuracy　is　very　high,　which　can　accurately　judge　the　abnormal　condition　of　the　nuclear　power　plant.　And　the　trained　abnormal　monitoring　model　can　independently　complete　the　abnormal　monitoring　task.　It　can　be　concluded　that　the　data-driven　abnormal　condition　monitoring　method　proposed　in　this　paper　can　accurately　and　timely　predict　the　abnormal　state　of　the　nuclear　power　plant.　©　2020,　Editorial　Board　of　Journal　of　Nuclear　Power　Engineering.　All　right　reserved.