Power　battery　is　the　energy　source　for　electric　vehicles.　It　is　of　great　significance　to　ensure　the　safety　and　reliability　of　electric　vehicles　by　accurately　predicting　power　battery　failures　and　identifying　their　fault　types.　Based　on　the　6-month　actual　vehicle　monitoring　data　of　10　pure　electric　vehicles,　16　feature　data　were　extracted　as　input　and　the　battery　fault　type　was　used　as　the　output.　A　double-layer　diagnosis　model　for　electric　vehicle　power　battery　faults　based　on　LightGBM　was　established　through　model　training　and　parameter　tuning.　The　upper-layer　model　was　used　to　determine　whether　the　vehicle　power　battery　is　fault.　The　lower-layer　model　diagnosed　and　analyzed　the　specific　fault　type.　Results　show　that　whether　the　electric　vehicle　power　battery　will　be　fault　can　be　correctly　predicted.　The　accuracy　of　diagnosing　the　fault　type　is　94.　05%　.　Meanwhile,　the　main　features　that　affect　the　failure　of　the　power　battery　are　screened　out　according　to　the　ranking　of　the　eigenvalues　of　the　model　results.　This　study　provides　an　approach　for　identifying　the　state　of　electric　vehicle　power　battery,　analyzing　the　fault　type　and　diagnosing　the　cause　of　the　fault.　©　2025　Beijing　University　of　Technology.　All　rights　reserved.