Abstract: Aiming at the problem of the over dependence of LF refining operation on manual experience，based on metallurgical mechanism，temperature model, alloy model, argon blowing model, and slag making model were developed on the 120 t LF of the Long Product Division of Maanshan Iron and Steel Co., Ltd.  And a control model was established based on the mutual unity of molten steel temperature，composition and slag.The big data technology and self-learning function were used to optimize the control model，the collaborative integration of various models and the whole process intelligent control of LF was realized，and good application results were achieved．The proportion of LF automatic control reached 80 %，the hit rate of endpoint target temperature ±5 ℃ was more than 95 %，the hit rate of narrow range of endpoint component (w(Si)±0.02 %、w(Mn)±0.02 %、w(S)±0.001 %、w(Al_s)±0.005 %) was above 97 %，the power consumption of LF refining was reduced by about 4 kWh/t，the refining treatment time was reduced by about 5 min，and the production efficiency and molten steel quality were improved．It played an important role in reducing cost and increasing efficiency during steelmaking process．

Key words: LF refining, intelligent control, model, integration, big data