Towards a multi-time-scale integrated solar forecast system

Compared to conventional fuel combustion systems, nuclear power stations and large hydroelectric plants, solar photovoltaic (PV) power generation is more variable. Thus, accurate forecasts of irradiance is an important tool for the cost-effective management of the undispatchable variability of solar power production. The need for reliable solar resource forecasts is becoming more important as increasing amounts of solar-generated electricity is incorporated into the electric grid. The objective of this project, sponsored by the California Energy Commission (CEC) under the auspices of the Public Interest Energy Research (PIER) Program, is to configure, demonstrate and validate a multi-time-scale solar forecasting system that provides short-term forecasts of solar-based generation. The integrated system look-ahead times will range from critical ramp identification period of minutes ahead to the frequency regulation time period of 1 hour ahead to the load-following time scale of approximately 6 hours ahead and ultimately to the day-ahead time frame that is important for generation unit commitment.

This project integrates Total Sky Imagers (TSI), NOAA geostationary satellites (GOES), and Numerical Weather Prediction (NWP) models into one seamless solar energy forecasting. A large operating solar PV plant in Nevada, U.S. was instrumented to capture the fine-scale temporal and spatial variability of the cloud-light environment. Two TSIs were deployed to collect and process cloud development, movement and sky cover near and around the PV plant for the purpose of forecasting intra-hour solar energy production. GOES visible satellite imagery provides data for short-term forecasts up to 6 hours ahead using a cloud vector advection scheme. Forecasts beyond 6 hours, up to several days ahead, are obtained from NWP models with a full suite of physics parameterization schemes (radiation, cloud microphysics, etc.) and a statistical post-processing algorithm to minimize systematic errors in the NWP forecasts. The multi-time-scale forecast system incorporates an Optimized Ensemble Algorithm (OEA) that constructs a composite deterministic or probabilistic forecast by statistically weighting the forecasts for the component methods by look-ahead time based on historical performance. Preliminary results from hindcasts or "forecasts in hindsight" for the individual components of the integrated forecast system at different time intervals (5 min, 15 min and 30 min) depending on the forecast horizon (0 to 30 min, 30 min to 5 hours, 6 to 48 hours) will be presented. The first phase of the validation process involves an objective evaluation of each individual forecasting method using standard metrics (e.g. normalized RMSE). Particular attention is given to the validation of 1-minute to 1-hour ramps as well as the intra-hour variability. The next phase will focus on the performance of the integrated multi-time-scale forecast system.