Physics-informed sparse deep neural networks for ultra-fast solar power forecasting and frequency response support in ancillary service markets

Abstract

The rising penetration of solar photovoltaic (PV) systems into the current power grid leads to large variations and uncertainties, which are key challenges to frequency stability and reliable auxiliary services. In this research, a novel Physics-Informed Sparse Deep Neural Network (PI-SDNN) framework is proposed for quick and accurate solar power forecasting specially built to enable fast frequency response (FFR) in auxiliary service markets. The proposed approach incorporates domain-specific physical constraints such as irradiance-power correlations, inverter dynamics and ramp-rate limitations, in a sparsity-promoting deep learning framework, which facilitates superior generalization, lower computational complexity and improved interpretability. The PI-SDNN model exploits structured physics-informed regularization to capture rich spatiotemporal features from high-resolution meteorological and historical generation data. The approach ensures fast inference suitable for real-time grid operations while preserving the forecasting accuracy in highly dynamic environmental scenarios. Furthermore, the forecasting findings are strongly related to a frequency response support module for proactive deployment and reserve allocation of solar-integrated ancillary services. Extensive simulations on real solar datasets demonstrate that the proposed framework outperforms classical deep learning and statistical models, yielding 25–40% improvement in forecasting accuracy metrics and significant reduction in computational latency. The integration of PI-SDNN with grid frequency management enhances the reaction speed and reduces frequency nadir variations during disturbances. The results show the potential of physics-informed sparse learning as a disruptive tool for reliable, scalable and commercially viable participation of solar in fast frequency response services, thus increasing the resilience and sustainability of future low-carbon power systems.