Water　quality　prediction　is　an　important　means　to　control　and　reduce　water　pollution.　Exist　water　quality　prediction　models　are　mainly　data　driven　and　only　depend　on　various　sensor　data.　At　present,　the　integration　of　machine　learning,　especially　deep　learning,　and　the　knowledge　graph　(KG)　has　become　a　research　hotspot.　Many　studies　have　proved　that　the　introduction　of　domain　knowledge　can　effectively　improve　the　data-driven　models.　This　paper　proposes　a　water　quality　prediction　model　integrating　KG　and　deep　adversarial　network.　The　KG　of　water　quality　is　extracted　by　the　joint　extraction　of　entities　and　relations,　and　introduced　into　prediction　modeling　as　prior　knowledge　for　parameter　importance　learning.　A　FreeAT-based　adversarial　learning　framework　is　combined　with　the　deep　prediction　model　to　improve　the　generalization　ability　of　model　in　a　few-slot　learning　scenario.　The　experimental　results　on　monitoring　data　from　the　Juhe　River　show　that　the　proposed　model　can　greatly　improve　the　robustness　of　the　model　and　reduce　the　prediction　error.