A Deep Autoencoder-Based Method for Detecting Mismatch Faults in Photovoltaic Modules

Abstract

Global initiatives to decrease reliance on fossil fuels and encourage a shift towards cleaner, more sustainable energy alternatives are gaining momentum. With technological progress, the efficiency and cost-effectiveness of photovoltaic (PV) modules are steadily improving, increasing their accessibility. However, weather and environmental factors can affect the effectiveness of PV modules, leading to fluctuations in energy production and a rapid decline in performance if not properly addressed. Therefore, early, accurate, and automatic fault detection is crucial for maintaining the reliability and efficiency of PV modules. Despite rapid advancements in machine learning methods, these approaches often rely on manually defined features, requiring expertise in selecting and extracting relevant features, and frequently fail to represent data comprehensively, resulting in incomplete information. This research presents a fault detection method for PV modules using deep autoencoder, leveraging current voltage curves and ambient conditions as input features. Deep autoencoder are applied for automatic feature extraction, dimensionality reduction, and effective fault detection within an unsupervised learning framework. This approach enhances detection accuracy and understanding of operational conditions and fault status of PV modules, aiming to prevent further damage, maintain optimal performance, and ensure system stability and reliability. Performance evaluation demonstrates that this method achieves high accuracy in fault detection for PV modules. To validate the proposed fault detection model, two other deep learning-based models namely, deep feedforward neural networks and deep belief networks are used for comparison.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint