Abstract
Wind and photovoltaic (PV) power output exhibits substantial temporal, spatial, and terrain-related variability, making sample partitioning critical for constructing representative output profiles. This study develops a unified framework for evaluating sample partitioning rules before clustering, rather than treating them as predetermined preprocessing conditions. The framework constructs 18 candidate schemes by combining different temporal, spatial, and terrain-based partitioning rules, and evaluates their suitability using six qualitative criteria and nine quantitative consistency measures through an integrated weighting and ranking approach. The selected scheme is subsequently used to organize daily output samples for clustering, from which hourly P50 characteristic profiles and P40–P60 empirical bands are extracted. For empirical validation, the framework is applied to wind and PV generation data from the Australian National Electricity Market over 2020–2025. The scheme combining monthly, county-level, and terrain-class partitioning, designated as A3, achieves the highest overall ranking among the 18 candidates. The resulting profiles characterize the intraday fluctuations in wind generation and the single-peaked daytime pattern of PV generation, while the quantile bands describe output dispersion within the dominant clusters. The methodological contribution lies in establishing a systematic pre-evaluation approach for sample partitioning rules and integrating it with subsequent output characterization, providing a structured basis for constructing condition-specific wind and PV output benchmarks.