Simulation-Based CNN-LSTM Fault Detection Framework for Grid-Connected Solar PV Systems

Abstract

The current study proposes an innovative simulation-based deep learning architecture for fault detection in solar photovoltaic (PV) systems connected to a grid system. A computerized PV system model was created in MATLAB Simulink to obtain data sets depending on irradiance level, temperature level, and faults. Data sets were normalized and divided into smaller segments using the sliding window technique and then fed to train the hybrid CNN-LSTM network. While CNN is responsible for spatial feature extraction in electric parameters, LSTM focuses on temporal feature extraction. The combination of CNN-LSTM model allows for reliable fault classification, including shading fault, open circuit fault, short circuit fault, and degradation fault. The CNN-LSTM network attained a classification accuracy of up to 97-98%, exhibiting high levels of robustness and versatility in fault detection in solar PV systems. The CNN-LSTM system design represents a breakthrough in the area of PV intelligent faults detection.