To　explore　the　application　of　deep　learning　theory　on　predicting　chloride　concentration,　3　150　groups　of　free　chloride　concentration　data　were　obtained　by　a　long-term　exposure　test　of　fly　ash　concrete　under　natural　tidal　environment.　A　multi-layer　perceptron　(MLP)　model　with　different　activation　functions　and　hidden　layers,　including　four　input　parameters,　namely　water-cement　ratio,　exposure　time,　fly　ash　content　and　penetration　depth,　was　established　to　predict　the　free　chloride　concentration　in　fly　ash　concrete.　Results　show　that　the　prediction　result　is　the　best　when　the　MLP　model　is　constructed　by　using　ReLu　function　with　four　hidden　layers.　Meanwhile,　the　selected　optimal　MLP　model　has　more　accurate　precision　when　predicting　the　free　chloride　concentration　based　on　the　not　measured　parameters,　compared　with　the　method　based　on　Fick’s　second　law.　Therefore,　the　MLP　model　has　the　advantages　of　high　precision　and　wide　application　scope,　which　can　be　used　as　a　new　method　for　predicting　free　chloride　concentration　of　concrete　under　chloride　environment.　©　2023　Beijing　University　of　Technology.　All　rights　reserved.