Pollution　of　Membrane　Bioreactor　(MBR)　remains　a　serious　issue　for　the　development　of　MBR　process.　An　early　warning　system　for　stifling　the　risks　of　membrane　pollution　is　of　prime　importance.　Misdiagnosis,　incorrectly　treatment　and　over　treatment　always　exist　in　traditional　early　warning　system,　which　cannot　satisfy　the　growing　requirements　of　these　wastewater　treatment　plants　(WWTPs).　Considering　all　the　factors　mentioned　above,　in　this　paper,　an　early　warning　system　based　on　data　and　knowledge　was　successfully　developed,　composed　of　forecasting　and　evaluation.　First,　a　scheme　was　designed　and　developed　for　the　early　warning　system.　Second,　a　data-driven　tbrecasting　model　was　proposed　as　an　important　part　of　this　system　based　on　the　theory　of　partial　least　squares　(PLS)　and　time　series　multi-step　prediction　method.　Finally,　the　early　warning　risk　level　was　evaluated　by　expert　knowledge　and　deep　belief　network　(DBN)　classifier,　meanwhile,　the　pollution　warning　levels　were　output　accordingly.　Experimental　results　demonstrate　that　both　the　accuracy　and　effectiveness　of　the　early　warning　system　are　greatly　superior.