Results archive for "Evidence channels dominate graded soft-set algebra in renewable-energy interval forecasting"

Results archive for a controlled comparison of fourteen uncertainty representations for interval forecasting of renewable-energy generation. The study asks whether the graded soft-set ladder — fuzzy, intuitionistic, Pythagorean, neutrosophic, plithogenic — earns its added degrees of freedom against the interval constructions practitioners already use.

This deposit contains the aggregated grid every number, table and figure in the article is emitted from: 37,800 scored entrant-cells with their interval scores, band–error faithfulness, coverage, width and CRPS; the fourteen-entrant comparison with bootstrap intervals; the pre-registered gate contrasts; the per-configuration point-forecast accuracy; the Holm-corrected pairwise model tests; and the nine figures as published. A README documents every column and every data source, and a MANIFEST.sha256 verifies inside the extracted archive.

Results cover 30 series across three public data sources — system-level German wind generation for the four transmission zones and the national total, turbine-level Kelmarsh SCADA, and a 24-city green-hydrogen production-potential proxy — at three horizons, each repeated under 30 recorded seeds: 2,700 paired comparison cells, 37,800 entrant-cells in total. Every entrant is scored inside the same cell on byte-identical inputs, around one shared fused point forecast and under one identical validation calibration, so the comparison is paired by construction and the representation is the only variable under test. The hydrogen series is a deterministic scenario proxy derived from reanalysis capacity factors, not a measured quantity.

Two results are worth naming. Under the canonical two-channel embedding the neutrosophic interval is algebraically identical to the aleatoric–epistemic quadrature interval, and the grid returns a discrepancy of exactly zero across all 2,700 cells, every metric and both regimes — which is what an identity must return, and what makes that gate row a check on the implementation rather than a discovery. Separately, the choice of evidence channel supplied to a band moves interval quality far further than the algebra layered over it, and expressive power does not order performance.

Not included: the implementation, which is available from the corresponding author on reasonable request, and the per-trial run directories, which are large and which every reported number aggregates away from. The raw input data are not redistributed; their sources, licences and SHA-256 checksums are recorded in the README so the processed series can be reconstructed and verified.