On the use of AI as a requirement for improved insolation forecasting accuracy to achieve optimized PV utilization

With the increase in data, the challenges of complex data, and the financial implications of improved insolation accuracy, it has become necessary to include targeted Artificial Intelligence & Machine Learning (AI/ML) for insolation forecasting. Forecasting is a crucial and cost-effective tool for integrating variable renewable resources into power systems. The ability to accurately forecast irradiance will facilitate increased PV adoption on the grid by making the intermittency less disrupting and allowing for better PV utilization, directly assisting in reduction of carbon energy sources.A key problem in solar forecasting is the intermittency of cloud cover, which often exhibits fractal properties and is still challenging to predict and adversely impacts solar farm output management. The physical models which cover weather can only provide a certain level of predictive accuracy and are particularly challenged by cloud forecasting. The key challenges are: limitations in the physical models, massive data, the need to make substantial simplifying estimations.We propose an approach and methodology which can enhance the predictive capabilities of insolation forecasting based on, and leveraging, a type of "bundled" approach which takes into account both the physical models, as well as the empirical mode determined by AI/ML, and exploiting sensor and satellite inputs. The novel aspect is to expand the AI/ML empirical dimension to achieve improved forecasting where the "non-physical-model" modes provide substantial input. We outline the specific methodology, how this is different from current modes, and how it can improve insolation forecasting. Specific examples will be provided and the benefits discussed.