Nowcasting, Doping and Discriminating Energy Generation of Photovoltaic Systems Using Regression Models of Adversarial Networks

This work assesses the ability of adversarial networks to forecast, enhance, and distinguish variations in output power generation within photovoltaic systems, extending from a previous classification-based model. We introduce a regression model that manipulates and discriminates deviations in output power to maintain a zero-sum balance. This model employs a Conditional Generative Adversarial Network (CGAN) comprising two main components: (i) a generator that predicts and simulates photovoltaic output power and (ii) a regression model that optimises the production process by integrating doping mechanisms and detecting discrepancies from expected output levels. We evaluated the effectiveness of our model using three real-world datasets. The results indicate that our model offers significant improvements in nowcasting, data enhancement, and deviation management, setting a robust foundation for ongoing research and advancement in renewable energy optimisation.