Improving performance and reliability of data driven predictive control through explainability.

Model Predictive Control (MPC) is a widely used control strategy that optimizes future control actions based on predictions of system behavior while respecting constraints. Traditionally, these predictions rely on explicit mathematical models of the system. However, for many modern applications such models are difficult to obtain, which has motivated the development of data‑driven predictive control (DPC). Instead of identifying a model, DPC directly uses previously recorded input–output trajectories to predict future system behavior and compute control actions.

While this approach can significantly simplify controller design, it also introduces new challenges. In realistic settings with noise or nonlinearities, purely data‑based predictions can become unreliable. To address this issue, many DPC methods include regularization terms that guide the controller toward meaningful predictions. However, the exact effect of these regularization choices is often unclear, and parameter tuning is frequently based on heuristics rather than systematic understanding.

The ExplainPRO project therefore aims to make regularized data‑driven predictive control more explainable and interpretable. By developing new mathematical analysis tools, we seek to characterize how different regularization strategies influence the predictions and closed‑loop behavior of these controllers. This will reveal hidden relations between existing approaches and help classify them according to their predictive behavior.

Building on these insights, the project will also develop improved control formulations whose regularization is chosen deliberately to enhance prediction quality, closed‑loop performance, and reliability. In this way, ExplainPRO aims to provide systematic design guidelines for data‑driven predictive control and contribute to more transparent and trustworthy data‑driven control methods. 

Related publications

M. Klädtke and M. Schulze Darup, On Data Usage and Predictive Behavior of Data-Driven Predictive Control With 1-Norm Regularization, IEEE Control Systems Letters, vol. 9, pp. 943-948, 2025. DOI: 10.1109/LCSYS.2025.3575436, Preprint: arXiv:2505.22307

M. Klädtke and M. Schulze Darup, Towards explainable data-driven predictive control with regularizations, at - Automatisierungstechnik, vol. 73, no. 6, pp. 365-382, 2025. DOI: 10.1515/auto-2024-0161, Preprint: arXiv:2503.21952

M. Klädtke and M. Schulze Darup, Implicit predictors in regularized data-driven predictive control, IEEE Control Systems Letters, vol. 7, pp. 2479-2484, 2023. DOI: 10.1007/LCSYS.2023.3285104, Preprint: arXiv:2307.10750

M. Klädtke, M. Schulze Darup, and Daniel E. Quevedo, Extending direct data-driven predictive control towards systems with finite control sets, 2024 European Control Conference (ECC), 3345-3350, 2024. DOI: 10.23919/ECC64448.2024, Preprint: arXiv:2404.02727

M. Klädtke and M. Schulze Darup, Towards a unifying framework for data-driven predictive control with quadratic regularization, Extended Abstract presented at the 26th International Symposium on Mathematical Theory of Networks and Systems MTNS 2024. Reprint: arXiv:2404.02721

M. Klädtke, D. Teichrib, N. Schlüter, and M. Schulze Darup, A deterministic view on explicit data-driven (M)PC, 61st IEEE Conference on Decision and Control (CDC), 499-504, 2022. DOI: 10.1109/CDC51059.2022.9993384, Preprint: arXiv:2206.07025