Modifier adaptation (MA) methods are iterative model-based real-time optimization (RTO) methods with the proven ability to reach, upon converge, the unknown optimal steady-state operating conditions of a plant despite plant-model mismatch and disturbances. So far, MA has been applied to small-scale but never-to the best of the authors' knowledge-to large-scale systems, the optimization of which being, in practice, a very difficult engineering problem. While standard MA uses plant measurements of the cost and constraints only, in this article, a new MA approach is proposed, namely Internal Modifier Adaptation (IMA), which allows the use of all available plant measurements leading to corrections at the level of the inner structure of the model. This article also provides a mathematical proof that IMA preserves the property of MA methods to reach the optimal inputs of the plant upon convergence. The application and the benefits of the proposed method are illustrated through two large-scale simulated case studies: (i) a steel-making plant, and (ii) the Tennessee Eastman challenge problem.