Abstract: The hot strip rolling process of strip steel is characterized by strong multivariable coupling, nonlinearity and time-varying behavior. Under unsteady conditions such as steel-grade transition, specification change, roll change and acceleration/deceleration, it remains difficult to satisfy the increasingly stringent requirements for three-dimensional geometry accuracy, strip shape and performance stability of high-end products. Focusing on data-driven modeling and high-precision three-dimensional geometric control in hot strip rolling, this paper reviews recent progress in machine-vision sensing, mechanism-data fusion modeling, dynamic digital twins, CPS-based coordinated control, and quality anomaly tracing with inverse process optimization. First, large-scale irregularly moving slab image acquisition, contour recognition and high-speed inter-stand strip deviation detection technologies were summarized, and their role in providing data support for closed-loop control of unsteady rolling processes was discussed. Second, the evolution of key models for rolling force, thickness, width and strip shape from conventional mechanism-based models to mechanism-data fusion digital twin models was analyzed. Third, the applications of hot rolling CPS systems in three-dimensional geometric control, including micro-tension and looper control, multi-stand thickness coordinated control, head warping and deviation control, and dynamic strip-shape optimization, were reviewed. Finally, full-process quality prediction, anomaly tracing and dynamic inverse optimization of process parameters were discussed. The reviewed studies indicate that the deep integration of data-driven methods and rolling mechanisms can effectively improve model-setting accuracy, dimensional control accuracy and quality stability under unsteady conditions, providing an important technical route for intelligent control of high-end hot-rolled strip production.