Towards Exact Temporal Aggregation of Time-Coupled Energy Storage Models via Active Constraint Set Identification and Machine Learning

TL;DR

This paper tackles exact temporal aggregation for power systems with energy storage time-coupling constraints. It introduces a theoretical framework that uses active constraint sets (ACSs) and dual information to disaggregate the full-scale model into independent submodels and then aggregates periods within submodels without loss of optimality, enabling parallel computation. To translate theory into practice, it couples this framework with machine learning: a feature-based classifier guides disaggregation, while hierarchical clustering aggregates within submodels, yielding substantial speed-ups with controlled accuracy. Numerical results across various energy storage and renewable configurations demonstrate near-perfect objective and decision-variable preservation under perfect information, and meaningful speed-ups when approximations are applied, highlighting the approach's potential for scalable, accurate long-horizon planning in energy systems.

Abstract

Time series aggregation (TSA) methods aim to construct temporally aggregated optimization models that accurately represent the output space of their full-scale counterparts while using a significantly reduced dimensionality in the input space. This paper presents the first approach that achieves an exact TSA of a full-scale power system model -- even in the presence of energy storage time-coupling constraints -- by leveraging active constraint sets and dual information. This advances the state of the art beyond existing TSA approaches, which typically cannot guarantee solution accuracy or rely on iterative procedures to determine the required number of representative periods. To bridge the gap between our theoretical analysis and their practical application, we employ machine learning approaches, i.e., classification and clustering, to inform TSA in models that co-schedule variable renewable energy sources and energy storage. Numerical results demonstrate substantially improved computational performance relative to the full-scale model, while maintaining high solution accuracy.

Figures (4)

• Figure 1: Conceptual 4-Step Argument and Accompanying Example for Exact TSA with ESSs.
• Figure 2: Disaggregation into Submodels (top) and Aggregation within Submodels (bottom) via Active Constraint Sets.
• Figure 3: Solution Error versus Number of Clusters.
• Figure 4: Maximum Speed-up versus Number of Clusters.