Deep learning for renewable energy forecasting: A taxonomy, and systematic literature review

In order to identify power production and demand in realtime for efficient and dependable management for diverse renewable energy systems, precise and intuitive renewable energy predictions are required. Deep learning can be exploited to handle a variety of operations and maintenance improvement challenges, as well as develop better methods and perspectives for medium-and long-term energy prediction. This paper provides a detailed literature and bibliometric review of deep learning models for effective renewable energy forecasting. To begin, data was gathered via the Web of Science (WoS) library to access a large amount of articles and journals. In WoS, a total of 276 publications were extracted, including five different types, including Articles (261), Reviews (13), Early Access (10), Proceedings Paper (5), Editorial Material (2), and Data Paper (1). Then, literature statistics analyzed top 10 productive country, author, institution, journal. Overall keyword analysis are explored and discussed from various aspects like most frequency keywords, keywords analysis by year, keywords co-occurrence analysis, keywords co-occurrence graph, topic evolution-accumulation, topic evolution weighted. In addition, literature statistics of renewable energy are evaluated for wind energy, solar energy, ocean energy, hydrogen energy. Deep learning models can be leveraged to anticipate underneath a variety of uncertainty arising from renewable energy sources that are fluctuating. Furthermore, the estimated prediction accuracy requirements are given, and the keywords analysis of the deep learning forecasting models iare demonstrated from the perspectives of SAE (Stacked AutoEncoder), DBN(Deep Belief Network), CNN(Convolutional Neural Networks), GAN(Generative Adversarial Networks), and RNN(Recurrent Neural Network). Due to various shifting weather conditions as well as other variables, the forecasting model responds differently by various types of datasets. Deep learning methods offer intriguing potential discoveries in the field of energy forecasting. The relevant aspects and suggestions for future research were highlighted in the conclusion to conquer the projected barriers.