Abstract: An important part of the iron-and-steel production process, converter steelmaking is the most widely used and efficient method of steelmaking in the world. Under the requirements of"China Manufacturing 2025, "ensuring intelligent steelmaking, improving smelting production efficiency, and reducing production cost are major concerns that should be addressed urgently in converter steelmaking. Owing to the complex thermodynamic and dynamic reactions in the converter smelting process, sublance control and traditional flue-gas analysis models have limitations that result in low prediction accuracy of the end-point carbon in converter smelting, thereby causing the main technical bottleneck in intelligent steelmaking. Therefore, a functional digital twin model of the steelmaking process based on flue-gas analysis was proposed. First, continuously monitored real-time data were obtained by flue gas analysis to observe the carbon and oxygen reaction state of molten steel in the converter. Then, according to various stages of the converter reaction, the functional data analysis method was used to establish the functional prediction models for the early and late stages of blowing. The greatest advantage of the method is that the model can automatically adjust the coefficient function according to the measured off-gas data by using a continuous functional curve to fit the complex dynamic reaction process. Therefore, the proposed model can accurately predict not only the normal smelting process but also the decarburization and carbon drawing process for the secondary scraping slag. An industrial experiment on a 260 t converter was conducted to prove that the functional digital twin model of the converter smelting process has good self-learning and self-adaptive ability and is robust to the abnormal smelting state. Furthermore, the model can predict the carbon content of the converter dynamically in the entire process and the end-point carbon content can reach 95% at ± 0. 02%. Using the predicted value of the carbon content to control the final blowing point through the functional digital twin model can effectively prevent overblowing or underblowing. More importantly, on the premise of guaranteeing the stability of raw material composition, temperature, weight, and other parameters, the model is expected to cancel the blown-off sampling step based on sublance. This feature can reduce the production cost while improving the product quality and production efficiency for a wide range of industrial applications.