Real-time　monitoring　of　surface　water　quality　is　an　intractable　problem.　A　Soft-sensor　method　based　on　fuzzy　neural　network　(FNN)　is　proposed　to　solve　this　problem　in　this　paper.　Firstly,　the　river　data　was　analyzed　by　principal　component　analysis　(PCA)　to　obtain　related　variables　such　as　dissolved　oxygen　(DO)　and　ammonia　nitrogen　(NH3-N).　Secondly,　a　multi-input　soft-sensor　method　based　on　FNN　is　designed.　The　training　data　is　preprocessed　by　Hierarchical　Clustering　and　K-means　algorithm　(H-K　algorithm),　which　improves　the　accuracy　of　the　soft-sensor　method.　Finally,　the　soft-sensor　method　is　packaged　and　applied　to　Beijing　Tonghui　River.　The　results　indicate　that　the　FNN　based　soft-sensor　can　predict　surface　water　quality　simultaneously　with　suitable　prediction　accuracy.　©　2019　Technical　Committee　on　Control　Theory,　Chinese　Association　of　Automation.