Smart Predict-Then-Optimize-Based Unit Commitment for Integrated Energy Systems

The unit commitment (UC) of the heat and electricity integrated energy systems (HE-IESs) is a fundamental issue in ensuring safe and economical operations, which requires predictions of renewable energy and load. In the traditional predict-then-optimize UC (TPO-UC) framework, the training process of the predictor is performed before and independently of the UC optimizer, potentially leading to suboptimal outcomes as it fails to account for the downstream UC optimizer’s concern about the prediction error. Considering this, this article proposes an improved UC framework for the HE-IES by integrating the concern of the UC optimizer into the predictor. First, we introduce the smart predict-then-optimize UC (SPO-UC) framework for the HE-IES, formulated as a bilevel mixed-integer linear programming (BiMILP) model. Second, we design a primal heuristic for SPO-UC to address the computational challenges, namely modified approximate reformulation and decomposition (MA-R&D), which is proved to converge to a stationary point of the equivalent single-level reformulation of BiMILP within finite iterations. Particularly, the proposed MA-R&D algorithm does not rely on the relatively complete response assumption and also accounts for the connection constraints in the SPO-UC model, which have not received attention in existing studies. Finally, case studies verify that the proposed SPO-UC outperforms the popular TPO-UC and the state-of-the-art SPO-UC methods in the HE-IES, and the proposed algorithm greatly improves computational efficiency.