Preprint: Unsupervised Learning-Based Operational Regime Discovery and Weather Sensitivity Analysis of a Residential Photovoltaic Battery System for Energy Management

The study introduces a label-free, data-driven framework designed to extract systemic operating archetypes from a behind-the-meter, grid-connected residential hybrid PV-battery system located in Yangon, Myanmar (22 kWp PV array, 50 kW Hybrid Inverter, 100 Ah battery bank). By evaluating 14 months of sub-hourly operational logs alongside co-located NASA POWER meteorological inputs, the methodology distinguishes four core operational regimes (C0, C1, C2, and C3) that transition dynamically from grid-dependence to complete self-sufficiency. Furthermore, the study deploys cluster-wise multivariate ordinary least squares (OLS) regression models to quantify how fluctuating weather components—specifically solar irradiance, ambient temperature, relative humidity, and wind speed—exert regime-specific impacts on daily PV energy yield.

Associated Digital Artifacts:

• Preprint DOI: https://doi.org/10.5281/zenodo.20507747 (This Upload)

• Dataset DOI: https://doi.org/10.5281/zenodo.18378566 (Contains the 14 months of high-resolution operational logs, 411 post-outlier engineered daily features, and the complete Python analytical pipeline script)

Keywords: Photovoltaic systems, Unsupervised learning, Clustering algorithms, Load profiling, Feature engineering, Renewable energy integration, Energy Management Systems, Weather sensitivity analysis.