Multi-scale solar radiation and photovoltaic power forecasting with machine learning algorithms in urban environment: A state-of-the-art review

Abstract

Solar energy has been rapidly utilized in urban environments owing to its significant potential to fulfill the energy demand. The precise forecasting of solar energy, including solar radiation and photovoltaic power forecasting, is crucial for effective energy utilization in cities. Currently, artificial intelligence algorithms, including machine learning (ML) and deep learning (DL) algorithms, have been widely utilized in the research of solar energy forecasting and have shown excellent performance. The different research scales lead to variations in the influencing factors and forecasting approaches used in solar energy prediction. Therefore, this paper provides a comprehensive review of solar radiation and photovoltaic power forecasting research using ML and DL algorithms from a multi-scale perspective, including the meso-scale, micro-scale, and building-scale forecasting in urban environments. The aim is to summarize the state-of-the-art progress and evaluate the solar forecasting effectiveness in diverse research scenarios. The characteristic analyses of ML- and DL-based solar forecasting approaches, prediction models, forecasting horizons, inputs and outputs were conducted for different research scales. This paper discusses the existing issues and future research directions for multiscale solar energy forecasting in cities. The analyses can help researchers and engineers in identifying suitable prediction algorithms and approaches. Consequently, this exploration can contribute to the efficient utilization of solar radiation and photovoltaic power generation in multi-scale urban environments.