Anomaly and Fault Classification and Prediction using Photovoltaic Digital Twins and Machine Learning based approaches

February 24, 2026

This thesis proposes a hybrid framework combining photovoltaic digital twins and Machine Learning for anomaly and fault classification and prediction. To address the scarcity of labelled real anomaly and fault data, digital twins were built using the PVlib Python library, simulating seven operational states: normal condition, three anomalies (soiling, shading, cracks), and three electrical faults (ground, arc, bypass diode). Synthetic daily and time series data were generated to train and evaluate multiple classifiers. Among tested algorithms, XGBoost demonstrated superior performance, achieving high classification accuracy, particularly for electrical faults. A probabilistic inference framework was applied to quantify prediction uncertainty and provide interpretable outputs for maintenance decisions. The trained model was successfully applied to real operational data for inference, showing plausible fault distributions. For prediction, a linear regression model was applied to forecast anomaly progression based on daily class probabilities. A sensitivity analysis identified Pareto-optimal parameter configurations, balancing early detection with prediction reliability. The results confirm that a digital twin-driven Machine Learning approach is a viable solution for predictive maintenance in photovoltaic systems, with XGBoost achieving 82.9% overall accuracy on the synthetic test set and up to 100% accuracy for distinct electrical faults such as ground and arc faults. The prediction framework successfully forecasted anomaly progression with success rates close to 100%, and an anticipation horizon of approximately 30 days. This work offers a practical pathway to enhance system reliability despite limited real fault data.

Authors in the community:

• 

  R

  Ricardo de Jesus Vicente Tavares

  ist1113368

Supervisors of this institution:

• 

  H

  Hugo Gabriel Valente Morais

  ist428549

• 

  A

  Amâncio Lucas de Sousa Pereira

  ist428524

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico