Decision Neutrality of Forecast Revisions in Microgrids with Photovoltaics and Battery Storage - Supplementary Material

This dataset is the supplementary material for the manuscript "Decision Neutrality of Forecast Revisions in Microgrids with Photovoltaics and Battery Storage" by Yun Xing (Hohai University), submitted to Applied Sciences (MDPI). Version 1.0, 25 October 2026.

Overview
The study asks when adopting a revised next-day forecast changes a battery dispatch decision, rather than only the predicted cost. It derives two checkable sufficient reset conditions under which old and revised forecasts produce next-day storage value functions that differ only by an additive constant. A fixed-input perturbation bound separates this local neutrality from realized rolling-policy performance. Five policies (FF, FS, SF, SS and DAY) are compared on 522 community-days from 270 measured households aggregated into nine synthetic communities, with three synthetic tariffs and two import regimes.

Key results
- Freezing next-day forecasts is numerically neutral under unlimited imports across the tested tariffs.
- Under a 30 kW import cap and time-of-use pricing, retaining midnight targets (DAY) lowers total operating cost by 3.105% relative to full updating (FF). Lower unmet-demand penalties account for 88.3% of the reduction; the import-bill reduction is 0.588%.
- Freezing both forecasts while replanning (SS) costs 0.054% more than FF under the same six-hour replanning schedule, separating forecast adoption from target retention.
- Under the same cap and tariff, freezing only the next-day forecast (FS) costs 0.188% more than FF.
- Matched-calendar comparisons show that the magnitude of update effects depends on the controller scheme.

Contents
1. Supporting_Results_and_Protocol.zip: verified reproducibility archive with 3,833 files, including fixed protocols, derived community inputs, customer-community mapping, analysis code, result tables, 15,660 policy-day records, 9,396 local terminal comparisons, 1,496 matched-calendar controller records, half-hour trajectories, figures and an internal manifest.
2. Supplementary_Guide.pdf: Supplementary Methods S1 and S2, extended uncertainty tables S1-S6, data dictionary notes and the run sequence.
3. README_ZENODO.txt: contents, usage order and scope notes.
4. LICENSE.txt: licensing note for code and derived data.
5. CITATION.cff: citation metadata.
6. zenodo_metadata.json: suggested metadata for this record.
7. SHA256SUMS.txt: SHA-256 checksums of the files in this archive.

Reproduction
Verify the files against SHA256SUMS.txt, extract Supporting_Results_and_Protocol.zip, and read README.txt, RUN_REPRODUCTION.txt and DATA_DICTIONARY.md before running anything. The recorded environment is Python 3.13.9 with NumPy 2.3.5, SciPy 1.16.3, pandas 2.3.3, Matplotlib 3.10.6 and openpyxl 3.1.5. Entry points cover the primary experiment, aggregation, availability sensitivity, diagnostics, both controller families, benchmarks and independent checks. The archive also supports a fresh single-group replay and a full audit without the author's original workspace.

Data provenance and scope
The household data are the Ausgrid 2012-2013 Solar Home Electricity Data, version 2, obtained from data.gov.au and the BetterGrids mirror. Raw provider records are not redistributed; provider URLs, version and SHA-256 are recorded in RUN_REPRODUCTION.txt and raw_source_manifest.json. Derived community inputs, results and trajectories are included. The archived results are the recorded reference outputs, not new observations. Communities, tariffs and batteries are synthetic scenario inputs, and the manuscript states the limits of the evidence.

License
Unless otherwise stated, the code and derived data in this package are provided under the Creative Commons Attribution 4.0 International License (CC BY 4.0). The original household data are not redistributed and remain subject to the provider terms.

Related manuscript
Submitted to Applied Sciences (MDPI). The manuscript DOI will be added to this record after publication.

Contact
Yun Xing, College of Computer Science and Software Engineering, Hohai University, Nanjing, China. Email: 2416010301@hhu.edu.cn. ORCID: 0009-0009-1746-5019.