Decision-Faithful Multi-Temporal Scenario Clustering for Renewable Power System Planning

Sustainable energy-system planning requires compact scenario sets that retain the renewable variability and operating stresses that determine investment, yet conventional typical-period methods can preserve average trajectories while removing these conditions. This paper develops a decision-faithful multi-temporal scenario clustering framework that evaluates representativeness through temporal structure and operating consequence. Hourly, daily, multi-day, and seasonal features describe source–load behavior, while a common screening fleet produces diagnostic signatures of cost, curtailment, adequacy, reserve pressure, ramping, storage-duration stress, transfer stress, scarcity price, and renewable-ramp interaction. A composite distance connects both representations to greedy facility location with seasonal and threshold-derived coverage, and assignment mass supplies scenario probabilities. The aggregate capacity-expansion benchmark comprises three years of hourly trajectories, a two-year design library, and a separate third-year dispatch evaluation. With 18 representative scenarios and design-calibrated (α,β)=(0.30,0.2), decision-signature error is 2.96%, compared with 4.87% for raw K-means, and capacity-vector error is 1.14%. Storage-boundary, screening-fleet, comparator, and exact-solver audits quantify the dependence on modeling choices. Across ten seeds from an independent unlabeled generator, signature error is 3.84±0.95% and capacity error is 15.38±6.66%. The computational experiments quantify representation and investment fidelity across the primary chronology, independent-generator cases, and sensitivity settings, and connect scenario reduction to renewable curtailment and adequacy under the planning model.