Abstract: Accurate regulation of the micro-differential pressure at converter mouth is crucial for improving converter gas recovery efficiency and realizing the green and low-carbon transformation of steel industry. To address the problems of response lag and insufficient stability in traditional proportional-integral-derivative control(PID)caused by transient strong disturbances such as oxygen valve switching and auxiliary material charging during the blowing process, this paper systematically summarizes the research progress of micro-differential pressure control models and classifies and evaluates the construction methods and application effects of models including fuzzy logic, artificial neural networks, Particle Swarm Optimization (PSO) and mechanism-data dual-driven models. Comparative analysis shows that existing technologies have difficulty balancing the requirements of high control precision and fast real-time response. Traditional PID control responds quickly but lacks working condition adaptability. Single intelligent algorithms represented by neural networks significantly improve the nonlinear fitting accuracy but suffer from dynamic response lag due to excessive computational load. Mechanism-data dual-driven models enhance their adaptability in complex scenarios by introducing physical constraints yet still face technical challenges such as difficult dynamic identification of model parameters, a sharp increase in system complexity and delay in underlying data transmission. This paper further conducts an in-depth analysis of the essential differences in control logic between the oxygen converter gas recovery(OG) and converter gas dry purification and recovery system (LT) and reveals the severe nonlinear regulation challenges of the LT caused by drastic flue gas temperature changes, electro-thermal-mechanical multi-physics coupling interference and actuator inertia lag. Based on this, it is proposed that future research should focus on developing adaptive fault-tolerant algorithms and constructing a digital twin sensing systems to achieve accurate characterization of the full-process dynamic characteristics and break through the control bottlenecks in the LT. This paper prospects the development trend of micro-differential pressure control models and point out that efforts should be made to improve the mechanistic interpretability of deep learning models, enhance multi-objective co-evolution capability and promote in-depth algorithm integration, which will provide solid theoretical support and technical guidance for the intelligent and green operation of the entire converter steelmaking process.