Long Short-Term Memory-Based Feedforward Neural Network Algorithm for Photovoltaic Fault Detection Under Irradiance Conditions

Monitoring and fault detection are crucial to increase the lifespan and dependability of photovoltaic (PV) systems. Numerous fault detection methods have been proposed in the literature over the last ten years. Certain fault detection techniques exclusively utilize data derived from I–V curves, whereas others depend on artificial neural network (ANN) methods to identify faults. In this study, open circuit faults, partial shading (PS) faults, short circuit faults, bypass diode faults (BF), mismatch faults (MF), line-to-line (L-L) faults, and inverter faults (IF) were considered as PV faults. However, addressing and resolving these issues in a timely and effective manner can potentially mitigate significant problems within PV systems. To resolve this issue, this article proposes a long short-term memory (LSTM)-based feedforward neural network (FFNN) algorithm using decision trees (DTs), support vector machines (SVMs), and linear regression (LR). In this approach, faults are categorized into eight distinct target classes because each fault considers two distinct conditions: the maximum and minimum power limits. Data were acquired under various irradiance levels to intentionally provoke the nonlinear characteristics of PV systems. The input data utilized consist of voltage-current characteristics. Compared to previous work, this approach requires only a limited dataset for learning and achieves enhanced accuracy in identifying and categorizing fault occurrences, particularly in scenarios involving high impedance and low mismatch levels. Even in difficult and complex situations, the results demonstrate the effectiveness and validity of the proposed strategy for classifying and detecting problems in PV arrays.