Seyed Masoud Parsa
Physics-informed machine learning (PIML) has emerged as a powerful paradigm to combine machine learning techniques with physical laws to enhance the accuracy, reliability and interpretability of machine learning (ML) models. This review presents recent advances in the application of PIML in renewable energy systems (RESs) through different aspects from macro-scale to micro-scale including wind energy (farm modeling, blade analysis, power output, spatiotemporal analysis, fault detection), solar energy (flat-plate collectors, photovoltaic, parabolic trough, heat transfer coefficients), biomass energy (kinetic reaction, syngas composition, biofuel production), geothermal (energy extraction, heat-exchanger performance), ocean current, and hybrid RESs. Following a discussion on different integration approaches of ML models and physics, the review critically analyzes diverse PIML approaches and architectures adopted in RESs by highlighting the progress in well-established technologies such as wind energy and emerging areas like ocean energy where PIML applications are still in early development. Moreover, insights on how a wide range of algorithms and optimization strategies have been adopted to solve domain-specific challenges within RESs using PIML are provided. Furthermore, current tools and open-source packages that support PIML implementations are also reviewed. Additionally, it outshines key criteria, indicators and guidelines for novel strategies to implement PIML in RESs. Finally, current challenges and future prospects are outlined. • Physics-informed machine learning for various renewable energy systems was summarized. • Wind energy is in the spotlight of applying PIML among all renewable energy systems. • PIML applications in solar, geothermal and biomass energy systems is in infant stage. • Neural network is the most applied ML method in all of renewable studies. • Indicator and criteria for selection PIML methods besides available Package and tools for implementing PIML was introduced.