Model predictive control is a popular advanced control method. The detection and diagnosis of model-plant-mismatches is an important task to ensure that an MPC is operating optimally and any potential model re-identification is targeted to only the sub-models that need it. Conventional detection methods directly use plant operating data for such purposes. Such methods fail in the presence of significant disturbances. A slow feature analysis data reconstruction is proposed to remove fast and typically irrelevant variations, extracting only those slow-varying and important components of the data to detect model-plant-mismatches. It is shown to have improved performance over a conventional method through both simulated and industrial case studies, and thus provide a more targeted selection of sub-models that need re-identification. © 2022 IEEE.