It would be impossible to discuss solar forecasting methods while not mentioning Artificial Intelligence. A lot of research is already focusing on the use of AI for solar forecasting [2,8,13,35] and also for decision-support tools in different domains of the electricity grid [3]. The wide range of different applications that have been applied successfully by researchers highlights the versatility of AI techniques. For example, AI has been used to develop energy bidding tools [36]; perform day-ahead solar forecasting [37], wind speed forecasting [35], solar radiation estimation [2,8,13,35], the monitoring of fields of PV systems [38], fault detection, and the diagnosis of wind energy systems [5]; and demand load predictions [3].

It should be noted that there is no official definition of what AI is [11] and what techniques and methods are considered Artificial Intelligence. Instead, AI is often used as an umbrella term to describe a wide variety of techniques, including but not limited to machine learning, supervised learning, optimization algorithms, pattern recognition techniques, and regression methods. One of the main strengths of what are generally considered AI techniques is that they are able to solve complex problems for which it is impossible to find explicit algorithms or mathematical solutions [7]. Often, this includes pattern recognition in large datasets in which the underlying principles or dependencies are very complex or unknown. The recent increase in the usage of AI techniques has been facilitated by a rapid increase in computational power over the last decades [39].

One of the most frequently used AI techniques in solar forecasting, as seen in Figure 1, and many other fields of research, is the Artificial Neural Network (ANN) [2,3,11,13,40]. The strength of ANNs is that they require a very low level of programming to solve a wide variety of complex problems. Specifically, nonlinear, stochastic, or mathematically ill-defined problems (e.g., pattern recognition or classification) are very well suited for ANNs [2,11,40,41]. Other popular techniques include the support vector machine [2,8,13], k-Nearest Neighbor algorithms [8,11], intelligent optimization algorithms [3], and Markov Chains [2,8,13]. Fuzzy Logic Control (FLC) has also been widely applied to control Solar PV systems and smart grids [3]. It makes it easy to use many input variables and to make use of the expert knowledge of human decision makers without the use of complex mathematical expressions [3,42].

A recent example of using AI methods for solar forecasting is the work conducted by Eseye et al. [43]. A data-driven approach was developed that employs a wavelet transform method, support vector machine, and particle swarm optimization to make predictions on the PV power output. The results were compared to seven other AI-based methods and proved to be competitive. This research also highlights the numerous methods already developed by using AI methods. Mishra and Palanisamy [44] developed a solar forecasting method built on Recurrent Neural Networks that was able to predict solar forecasts over a wide range of time horizons ranging from intrahour, hourly, to day-ahead scales using real-time inputs. Another example that showcases the possibilities of AI within solar forecasting is the method developed by Ge et al. [45]. They developed a method that only uses empirical data and AI, thus excluding the use of any physical model or empirical relationship while still being able to achieve similar results to more classical methods. Another method that has been gaining interest for nowcasting is the General Adversarial Network (GAN) method, which has already proven to be able to forecast precipitation with high precision [46], improve time series Satellite Image prediction [47], and perform sky image forecasting [48]. For further reading on AI techniques, refs. [8,11] are recommended.