Project Name: HAIMOS Ensemble Forecasts for Intra-day and Day-Ahead GHI, DNI and Ramps
Funding Opportunity: Solar Forecasting 2
SETO Subprogram: Systems Integration
Location: San Diego, CA
SETO Award Amount: $1,316,203
Awardee Cost Share:  $162,500
Principal Investigator: Carlos F. M. Coimbra

This project will develop the Hybrid Adaptive Input Model Objective Selection (HAIMOS) ensemble model for solar irradiance forecasting. HAIMOS is a physics-based and data-driven model that forecasts both direct normal irradiance (DNI) and global horizontal irradiance (GHI) for horizons up to 72 hours in advance. One of the key gaps in these technologies is the lack of accurate solar forecasts for DNI and inaccurate forecasting of large, sudden changes in irradiance, known as irradiance ramps. This project aims to develop a forecast accuracy that is considerably higher than that of the persistence or baseline forecast, across a wide range of time horizons for both GHI and DNI.

APPROACH

This model combines numerical weather prediction forecasts, deterministic and physics-based algorithms, and new-generation cloud-cover products like high-resolution rapid refresh satellite images and large eddy simulations (LES), which are mathematical models that simulate atmospheric air currents. Data preprocessing, input selection, and error metrics, will be optimized to reduce DNI and GHI forecast error and improve the prediction of ramp onset. The research team will focus on improving cloud identification and forecasting of both cloud cover and cloud optical depth or opacity. They will then develop improved algorithms to identify whether a cloud exists in a particular location, what type it is, and the value of certain parameters that describe its behavior. The final forecasts will adaptively take into account the advanced imagery and various other inputs and data-streams.

INNOVATIONS

HAIMOS combines the latest innovations in machine-learning algorithms with detailed physics-based models for forecasting cloud cover and cloud optical depth. Another innovation is the integration of information derived from the high-resolution sensors in the new geostationary satellites and cloud cover simulations from LES. Both of these technologies enable the spatial-temporal sensing and modeling of clouds—data collected across time and space—at much higher resolutions than previously available.