Exploring deep learning methods for solar photovoltaic power output forecasting: A review

Abstract

The rise of distributed energy resources stems from reliance on carbon-intensive energy and climate concerns. While photovoltaic solar energy leads in modern grids, its intermittent nature and weather variability challenge reliability and efficiency. Photovoltaic power output forecasting ensures a stable power supply by mitigating weather-induced disruptions. Thus, this review paper investigates the transformative impact of Deep Learning (DL) on photovoltaic power output forecasting. Leveraging the extensive data generated by smart meters, DL has shown unprecedented potential to outperform traditional forecasting models. The primary purpose of this research is to systematically analyze and compare mainstream DL-based forecasting techniques, uncovering their respective strengths and limitations. Least explored techniques such as deep transfer learning, big data DL, federated learning, probabilistic models, deterministic models, and hybrid architectures in forecasting are explored which have distinct advantages in processing large-scale multi-source data to deliver more accuracy. Covering research from 2019 to 2023, this study aims to capture the latest developments and ensure relevance to ongoing trends. Nearly 200 journals were acquired for this review paper using a systematic protocol. Among the DL methods, Autoencoder-Long Short-Term Memory outperformed its counterparts, achieving an impressive R² score of 99.98%. Moreover, the major conclusion underscores that DL offers a promising pathway for advancing PV forecasting, with future opportunities to address identified gaps and emerging challenges. This analysis serves as a comprehensive guide to stakeholders, illuminating the unique capabilities of DL in driving the next generation of solar power forecasting solutions.