気候帯を横断したUMAP–HDBSCANとSHAPによる全球ソーラー気象レジーム同定のための説明可能AI

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:気候帯横断でソーラー気象レジームを説明可能に同定する手法は、再エネ資源評価研究に新たな分析枠組みを提供する。

🏢実務担当者:レジーム別の最適傾斜角や設計示唆は、PVシステム設計・発電量予測・系統統合計画に直接活用できる。

🏛政策担当者:気候適応型の再エネ計画立案や、地域別の太陽光導入支援策の設計に科学的根拠を提供する。

Solar photovoltaic (PV) generation is shaped by recurring combinations of irradiance and atmospheric conditions, yet many machine-learning approaches treat weather observations independently and overlook these repeatable patterns. This preprint develops an explainable unsupervised learning framework to identify and interpret solar-weather regimes across global climate zones. The study analyses 67,287 hourly meteorological observations from 17 cities representing tropical, subtropical, temperate, and cold climates using a three-stage analytical pipeline: Uniform Manifold Approximation and Projection (UMAP) for nonlinear dimensionality reduction, Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) for regime discovery, and an XGBoost proxy classifier with Shapley Additive Explanations (SHAP) for transparent interpretation. Key Findings: Six distinct solar-weather regimes are identified, each characterised by coherent meteorological signatures and systematic variations. SHAP analysis indicates that irradiance partitioning (direct normal irradiance and diffuse horizontal irradiance) together with relative humidity and air temperature are the dominant discriminators of regime membership. Regime-specific photovoltaic design implications are substantial: optimal tilt angles vary by more than 40°, ranging from shallow configurations (≈7–8°) in diffuse-dominated regimes to steep configurations (>50°) under low-sun conditions. Novel Contributions: - Regime-based characterisation of solar-weather variability through joint analysis of irradiance components and meteorological variables - Adaptation of weather-regime concepts from climatology to solar energy applications - Explainable interpretation of unsupervised regime discovery via transparent proxy modelling with SHAP - Consistent multi-climate comparative analysis across 17 globally distributed cities spanning all major climate zones - Direct linkage between atmospheric regimes and photovoltaic system performance implications The proposed framework provides a data-driven basis for solar resource characterisation, actionable regime-level insights, climate-aware photovoltaic system design, and renewable energy planning across diverse climatic settings. Methodology: The analysis employs reanalysis-based hourly meteorological data (temperature, humidity, wind speed, pressure) and solar irradiance variables (GHI, DNI, DHI) from 17 cities. All data were screened to remove missing entries and non-physical values. Variables were standardised prior to unsupervised learning. UMAP enables nonlinear dimensionality reduction whilst preserving neighbourhood structure. HDBSCAN identifies dense regions corresponding to recurrent regimes without requiring predefined cluster numbers. XGBoost with SHAP provides transparent feature attribution and regime interpretation. Audience & Applications: This work is relevant to: - Solar resource engineers and researchers - Photovoltaic system designers and energy planners - Climate and atmospheric scientists - Renewable energy policy makers - Machine learning practitioners in energy systems The framework supports improved solar resource assessment, climate-adaptive PV design, regime-informed forecasting, and grid integration strategies. Keywords: Solar photovoltaic; solar-weather regimes; unsupervised learning; explainable AI; UMAP; HDBSCAN; SHAP; climate zones; renewable energy; irradiance; machine learning Funding: This work was supported by the Faculty of Electrical Engineering and Computer Science, Technical University of Ostrava, under the project "Research Platform for Digital Transformation and Society 5.0" (CZ.02.01.01/00/23_021/0012599), funded by the European Regional Development Fund within the Jan Amos Komenský Operational Program. Authors: Ohn Zin Lin¹*, Hnin Yee Aye¹, Paing Hein Soe², Libor Štěpanec¹, Eftichios Koutroulis³, Dagmar Juchelkova¹ ¹ Department of Applied Electronics, Faculty of Electrical Engineering and Computer Science, Technical University of Ostrava, Czech Republic ² Department of Mechanical and Industrial Engineering, Northeastern University, Boston, USA ³ School of Electrical and Computer Engineering, Technical University of Crete, Greece Citation: Lin, O.Z., Aye, H.Y., Soe, P.H., Štěpanec, L., Koutroulis, E., & Juchelkova, D. (2024). Explainable AI for identifying global solar weather regimes using UMAP–HDBSCAN and SHAP across climate zones. Preprint SSRN-7148141. Corresponding Author: Ohn Zin Lin ([email protected]) Department of Applied Electronics, Faculty of Electrical Engineering and Computer Science Technical University of Ostrava, Czech Republic

• Zenodo https://zenodo.org/records/22892752first seen 2026-09-23 04:28:33 · last seen 2026-09-25 04:12:32