Hybrid Forecasting Systems for Integration of Large-Scale Solar Power Plants

Increasing penetration of weather-dependent renewable power into the electric grid requires strategies for coping with generation variability. Of the existing solutions, renewable energy forecasting is the cheapest and most readily deployable for both distributed generation and utility-scale central station power plants. Our group at UC San Diego has developed a number of specific forecasting systems that combine physics-based energy meteorology models with image analysis from both ground and geo-stationary satellite sources, all interlaced by machine learning techniques. These hybrid methods allow us to predict not only the position and optical depth of clouds, but also their movement over solar fields, and consequently the potential for solar power generation at the ground level. In this talk I will discuss some of the previous success stories in the field of energy meteorology, but also the ongoing efforts to improve forecast fidelity with the next generation of solar forecasting engines.

Speaker: Carlos Coimbra, UC San Diego