Optimizing photovoltaic output performance prediction: a deep learning approach with LSTM neural networks and Adam optimizer / Syasya Nadhirah Hamedon, Juliana Johari and Fazlina Ahmat Ruslan

This study introduces an innovative approach to optimizing photovoltaic (PV) output performance prediction through Deep Learning, specifically employing Long Short-Term Memory (LSTM) networks and the Adaptive Moment Estimation (Adam) optimizer. The research is carried out using MATLAB R2023a, and the dataset is publicly accessible from the Solcast website. Performance evaluation utilizing key indicators such as Loss Metrics and Root Mean Squared Error (RMSE) highlights the potential of this Adam-optimized LSTM method to significantly enhance the accuracy of PV performance prediction. The analysis explores the impact of learning rates, evaluating fixed rates (0.0001, 0.001, 0.01) and a dynamic transition (0.01 to 0.001) over 10 epochs. Notably, a learning rate of 0.01 demonstrates substantial improvements, achieving lower errors and consistently low losses, indicating a highly accurate and well-fitted model. The unexpected adaptability observed during a dynamic learning rate transition further highlights the model's potential. The study presents a comprehensive analysis of LSTMs and Adam optimizer for PV output performance prediction and provides valuable insights for researchers seeking optimal learning rates to develop robust and effective PV performance prediction models.