Sensitivity analysis of sampling and clustering techniques in expansion planning models

Short and long-term power system planning models are becoming more complex in order to capture current and future market characteristics comprising more variability, uncertainty, and integration of geographically spread market areas. Dimension reduction methods can be used to keep the planning models tractable, e.g. time series sampling and clustering, but they represent a trade-off between model complexity and level of detail. The accuracy of dimension reduction methods can be measured both in terms of raw data processing and model output metrics, where the latter reveals how well a sampling technique fits that particular model instance. In this study, the robustness of several sampling and clustering techniques is quantified with different model instances by independently varying model parameters, such as e.g. the marginal cost of generation. As the obtained findings indicate that the performance of the considered techniques is, indeed, model-dependent, more insight into the performance of common dimension reduction techniques in power system planning applications is provided. The results are illustrated by a case study of the North Sea Offshore Grid (NSOG) for the scenario year 2030, using a bi-level mixed-integer linear optimization program. All things considered, systematic sampling and moment matching are shown to give the most robust results from the sensitivity analysis.