Predictive modeling and anomaly detection in solar PV inverters using machine learning

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60076658%3A12510%2F25%3A43910136" target="_blank" >RIV/60076658:12510/25:43910136 - isvavai.cz</a>

Result on the web

<a href="https://www.sciencedirect.com/science/article/pii/S2590123025043920" target="_blank" >https://www.sciencedirect.com/science/article/pii/S2590123025043920</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1016/j.rineng.2025.108348" target="_blank" >10.1016/j.rineng.2025.108348</a>

Result language

angličtina

Original language name

Predictive modeling and anomaly detection in solar PV inverters using machine learning

Original language description

The operational stability of photovoltaic (PV) systems is critical to the success of distributed renewable energy integration. This study presents a machine learning-driven framework for performance modeling, anomaly detection, and classification of inverter output in a grid-connected PV installation. Using high-resolution data collected from 30 kW and 40 kW inverters over one month, we applied supervised learning techniques to predict active power output, categorize production levels, and detect deviations from expected behavior. A Random Forest Regressor achieved high fidelity in power prediction (R2 = 0.995, MAE = 0.12 kW), while classification models categorized output levels with 100 % accuracy under static conditions. Anomaly detection using Z-score analysis identified significant outliers, particularly during high-production intervals. However, one-hour-ahead classification revealed substantial drops in predictive performance (accuracy = 36.4 %), highlighting the inherent difficulty of forecasting under variable environmental conditions. Feature importance analysis underscored the dominant role of inverter input power and phase currents in prediction accuracy. These findings demonstrate the utility of interpretable, data-driven models for real-time diagnostics and forecasting in PV systems and lay the groundwork for scalable smart monitoring infrastructures. Unlike prior studies that rely on meteorological inputs, this work uses only inverter and grid-side electrical measurements, demonstrating that interpretable models can provide actionable insights even in data-limited scenarios. The proposed framework offers a practical foundation for early-warning diagnostics and intelligent maintenance scheduling, enabling more resilient PV operations. Its integration of interpretable machine learning with real-time monitoring distinguishes it from conventional rule-based supervision methods.

Czech name

—

Czech description

—

Type

J_imp - Article in a specialist periodical, which is included in the Web of Science database

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Publication year

2025

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Name of the periodical

Results in Engineering

ISSN

2590-1230

e-ISSN

2590-1230

Volume of the periodical

28

Issue of the periodical within the volume

December 2025

Country of publishing house

NL - THE KINGDOM OF THE NETHERLANDS

Number of pages

13

Pages from-to

1-13

UT code for WoS article

001633402000009

EID of the result in the Scopus database

2-s2.0-105023579529