Sludge　bulking　is　a　common　abnormal　condition　in　municipal　wastewater　treatment　process　(WWTP).　It　is　difficult　for　WWTP　to　effectively　achieve　high-precision　feature　extraction　since　it　has　complex　reactions,　many　influencing　factors,　and　strong　coupling　of　factors.　In　this　paper,　a　multi-kernel　combined　principal　component　analysis　(MKPCA)　method　for　extracting　characteristic　variables　of　sludge　bulking　based　on　multi-innovation　random　gradient　is　proposed.　Firstly,　based　on　the　nonlinear　characteristics　of　kernel　functions　and　the　advantages　of　adaptability　of　different　kernel　functions,　a　multi-kernel　combination　mechanism　is　designed.　Then,　a　principal　component　analysis　method　based　on　multi-kernel　combination　is　proposed.　Secondly,　a　multi-innovation　random　gradient　identification　method　is　designed,　which　introduces　a　sliding　window　mechanism　to　update　the　structure　and　parameters　of　kernel　functions　with　multiple　time　data.　In　addition　to　ensuring　the　identification　accuracy,　the　variable　feature　extraction　effect　of　sludge　bulking　process　can　be　improved.　Finally,　the　method　is　tested　with　actual　data　from　wastewater　treatment　plant.　The　results　show　that　the　proposed　method　has　a　better　feature　extraction　effect.