Uncalibrated AI models limit the transferability of probabilistic photovoltaic forecasts

Probabilistic forecasts of Photovoltaic (PV) power are useful only if their intervals remain calibrated at new sites, in new seasons, and with the information available when forecasts are issued. This review asks whether AI models deliver such forecasts. Journal articles from 2022 to 2026 were retrieved from OpenAIRE and Crossref. A bibliometric map of 169 articles was combined with an abstract-level evidence map of 101 core studies, technical appraisal of eight full texts, and a structured synthesis of 13 matched proper-score contrasts. Output rose from 27 articles in 2022 to 66 in the partial year 2026, with conformal prediction and transfer learning appearing only from 2024 and foundation models in 2026. Among 77 core abstracts, 43% stated a horizon class or lead time, 22% a proper scoring rule, 34% interval coverage, 3% conditional coverage, 5% a new site or transfer, and none a tropical site. Within studies, AI models lowered the continuous ranked probability score or interval score against every comparator, by a median of 27% against the strongest baseline in the most complete study and by 12% to 13% in two others, always in-distribution. Only four of the eight appraised studies demonstrably used information available at issue time, and the only comparison with an operational day-ahead forecast favored the operational forecast. Wherever coverage was reported numerically by regime, nominal 90% intervals undercovered, falling to 76% in daylight hours. The evidence supports AI for more accurate PV forecast distributions in-distribution, not yet for calibrated and transferable forecasts under shift.

Authors

Publication Details

Journal
Journal of Engineering and High Technology
Published
2026-11-15
DOI
https://doi.org/10.55942/jeht.v2i2.2370
Primary Topic
Solar Radiation and Photovoltaics
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Uncalibrated AI models limit the transferability of probabilistic photovoltaic forecasts

Muhammad Zaki Rasyadi
Journal of Engineering and High Technology
Solar Radiation and Photovoltaics
article

Uncalibrated AI models limit the transferability of probabilistic photovoltaic forecasts

Muhammad Zaki Rasyadi
article en

Abstract

Probabilistic forecasts of Photovoltaic (PV) power are useful only if their intervals remain calibrated at new sites, in new seasons, and with the information available when forecasts are issued. This review asks whether AI models deliver such forecasts. Journal articles from 2022 to 2026 were retrieved from OpenAIRE and Crossref. A bibliometric map of 169 articles was combined with an abstract-level evidence map of 101 core studies, technical appraisal of eight full texts, and a structured synthesis of 13 matched proper-score contrasts. Output rose from 27 articles in 2022 to 66 in the partial year 2026, with conformal prediction and transfer learning appearing only from 2024 and foundation models in 2026. Among 77 core abstracts, 43% stated a horizon class or lead time, 22% a proper scoring rule, 34% interval coverage, 3% conditional coverage, 5% a new site or transfer, and none a tropical site. Within studies, AI models lowered the continuous ranked probability score or interval score against every comparator, by a median of 27% against the strongest baseline in the most complete study and by 12% to 13% in two others, always in-distribution. Only four of the eight appraised studies demonstrably used information available at issue time, and the only comparison with an operational day-ahead forecast favored the operational forecast. Wherever coverage was reported numerically by regime, nominal 90% intervals undercovered, falling to 76% in daylight hours. The evidence supports AI for more accurate PV forecast distributions in-distribution, not yet for calibrated and transferable forecasts under shift.

Journal of Engineering and High TechnologyVol. 2(2)
Solar Radiation and Photovoltaics
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon