Given　the　numerous　studies　on　how　to　make　buildings　smarter　through　smart　devices,　there　is　a　need　for　more　research　on　retrofitting　large　buildings　with　legacy　equipment　and　systems.　We　propose　a　deep　learning　based　cyber-physical　control　solution　for　smart　retrofitting　of　buildings　in　service.　Specifically,　our　solution　first　collects　real-time　operational　data　without　interrupting　the　operation　of　the　original　control　system,　and　then　utilizes　the　LSTM-CNN　prediction　model　to　learn　the　multi-scene　control　strategy　embedded　in　the　historical　data.　Based　on　this　model,　intelligent　retrofitting　of　many　large　buildings　can　be　realized,　thus　significantly　reducing　the　manpower　required　for　manual　operation　and　maintenance.　The　model　was　experimented　using　real　data　from　the　HVAC　system　of　the　T3　terminal　building　of　Beijing　Capital　International　Airport,　and　the　results　show　that　the　proposed　solution　can　be　used　in　conjunction　with　existing　equipment　to　accomplish　the　smart　retrofit　and　can　effectively　replace　manual　control.　©　2024　IEEE.