2296. What Are We Clustering For? Establishing Metrics and Performance Guarantees for Time Series Aggregation in Net-Zero Power System Optimization
Invited abstract in session WC-3: Energy planning and policy, stream Energy and Sustainability.
Wednesday, 13:30-15:00Room: H5
Authors (first author is the speaker)
| 1. | Luca Santosuosso
|
| Institute of Electricity Economics and Energy Innovation, TU Graz | |
| 2. | Beltrán Castro Gómez
|
| Graz University of Technology | |
| 3. | Bettina Klinz
|
| Institut für Optimierung und Diskrete Mathematik, TU Graz | |
| 4. | Sonja Wogrin
|
| Graz University of Technology |
Abstract
The transition toward net-zero power systems necessitates high-fidelity optimization models that rely on high-resolution input time series of renewable generation, demand, and other exogenous factors. However, this leads to significant computational challenges, often rendering such models intractable. Time Series Aggregation (TSA) alleviates this issue by reducing temporal complexity through representative period selection via clustering. Yet, existing TSA methods have notable limitations. A priori methods focus solely on preserving the statistical properties of the input data, resulting in heuristics that lack theoretical guarantees on the accuracy of the aggregated model’s output. In contrast, a posteriori methods directly aim to preserve the accuracy of the aggregated model, typically using the objective function as the sole evaluation metric. However, this may not align with the goals of all modelers, who often seek accuracy in specific decisions, such as technology- or region-specific investments, not directly captured by the objective function. This study revisits the TSA problem by tackling a key research question: What are we clustering for? First, we advance beyond the heuristic nature of a priori methods by introducing theoretically validated bounds on the objective function error for aggregated models in time-coupled mixed-integer power system investment planning problems. Second, we propose alternative evaluation metrics that reflect diverse modeling goals, beyond the objective function. Finally, we extend the theoretical guarantees of the proposed method to these metrics. Numerical results show the computational efficiency of the proposed TSA method and underscore the importance of metric selection to align with the specific goals of distinct modelers.
Keywords
- Energy Policy and Planning
- Computational Complexity
- Mixed-Integer Programming
Status: accepted
Back to the list of papers