2248. Decision-Based vs. Input-Driven Timeseries Clustering for Stochastic Energy System Design Optimization
Invited abstract in session TC-44: Approaches for handling computational complexity, stream Energy Economics & Management.
Tuesday, 12:30-14:00Room: Newlyn 1.01
Authors (first author is the speaker)
| 1. | Boyung Jürgens
|
| Institute of Technical Thermodynamics, RWTH Aachen University | |
| 2. | Andrew Fulwider
|
| Institute of Technical Thermodynamics, RWTH Aachen University | |
| 3. | Hendrik Schricker
|
| Institute of Technical Thermodynamics, RWTH Aachen University | |
| 4. | Niklas von der Assen
|
| Institute of Technical Thermodynamics |
Abstract
Two-stage stochastic programming is widely used for energy system design optimization under uncertainty but can face an exponential increase of computational demand with the number of scenarios. Clustering reduces the number of scenarios: Input-driven clustering (IDC) groups scenarios by similar inputs, e.g., demand projections. Decision-based clustering (DBC) groups scenarios by similar first-stage (i.e., design) decisions. In energy system design optimization, scenario variability often arises from timeseries data. Performance comparisons of IDC and DBC in energy system optimization frequently relied on timeseries generated by scaling reference patterns, which limits insights for varying timeseries shapes.
In this study, we apply IDC and DBC to energy system design optimization with scenarios characterized by timeseries of varying shapes. The case study models a sector-coupled industrial energy system as a linear program. Uncertainty stems from up to seven timeseries, describing energy demands, energy carrier prices, and availabilities of renewables. We assess the performance of IDC and DBC across varying numbers of uncertain timeseries, scenarios, and clusters.
Our findings imply that DBC can be more robust against uncertain energy demands than IDC. Our case study shows that DBC can remain more robust when varying the number of uncertain timeseries but is sensitive to the method used to group first-stage decisions, while IDC is computationally efficient but less robust.
Keywords
- Energy Policy and Planning
- Programming, Linear
- Stochastic Optimization
Status: accepted
Back to the list of papers