Enhanced photovoltaic output prediction through attention-aware transfer learning and image denoising

The purpose of this study is to develop a comprehensive framework that highly enhances the accuracy of photovoltaic (PV) output prediction using deep machine learning. As our approach, the denoising images with Generative Adversarial Networks for preprocessing raw data, employing autoencoders to uncover meaningful features and feature selection based on Firefly Algorithm for identifying most important predictors are indispensable in our method. In order to confirm that our proposed method was effective, extensive experiments were conducted which compared it with conventional approaches including Convolutional Neural Networks (CNN), Long Short-Term Memory networks, Autoregressive Integrated Moving Average and others. Consequently, this model performs better than other models on these datasets by having low mean error rates in both overall accuracy as well as index performance variability across evaluation metrics. Also, we have several advantages of our proposed framework that include increased correctness, insensitivity to noise or alteration of data and the ability to adapt different types of PV system architectures such as. It is a renewable energy source from the sun which is an easily accessible and long lasting one that can be a viable solution for world energy scarcity forever. Evaluating these outcomes proves that our approach can modify PV-power predictions, therefore making this solar technology more efficient and sustainable. This research helps in improving renewable energy technologies and supporting shift to more resilient infrastructure for power.