Data-Based Robust Model Predictive Control Under Conditional Uncertainty

In this work, a novel data-driven robust model predictive control (RMPC) framework is outlined for optimal operations and control of energy systems, where uncertainty in predictions of energy intensities enters into the process in an additive manner. However, the distribution of prediction errors may be time-varying and depend on other external variables. To appropriately describe the distribution of uncertainty, a novel concept of conditional uncertainty as well as the conditional uncertainty set is proposed, which disentangles the dependence of distribution on external variables and hence reduces the conservatism. In general, the conditional uncertainty set can be modelled by integrating domain-specific knowledge and data collected from previous experience. An example arising from agricultural irrigation control is presented to illustrate of the effectiveness of the proposed methodology.