Optimizing PV Array Performance: A² LSTM for Anomaly Detection and Predictive Maintenance based on Machine Learning

Photovoltaic (PV) energy is considered one of the most promising renewable sources. Detecting and monitoring faults in PV systems ensures optimal efficiency and prevents safety and hazards. Predictive maintenance (PdM) is the prominent anomaly prediction strategy that predicts health conditions with machine learning (ML) algorithms. However, existing algorithms overlook the importance of attribute consideration and fail to account for temporal dependence in final results. To address such issues, this paper proposes the implementation of Attribute Attention (A2)-based Long short-term memory (LSTM) for general PdM framework based on clustering and anomaly detection in PV array data. The A2-LSTM model is complemented by an unsupervised K Means clustering technique to identify patterns within the data. The attention mechanism in the attribute attention-based LSTM model is used to identify the most relevant attributes in PV array electrical data for each cluster, allowing the model to focus on the information that is most pertinent to predicting the behavior of the PV arrays within that cluster. The results indicate that the proposed model identified anomalies in the predicted data of the PV array more accurately. The proposed model could help the plant operator perform Remaining Useful Life (RUL) for PdM to carry out PV array maintenance. To the best of our knowledge, the A2 method is not been used for the PdM problem of PV plants.