As solar energy becomes a bigger part of the generation mix, grid operators need better visibility of how much solar power the system is producing. Then they can optimally dispatch solar and other energy sources to balance the generation and demand. Improving solar forecasts will allow the electric grid to be more flexible and adapt to changing conditions while helping to minimize disruptions and the overall cost of operation. Research is improving the accuracy of these predictions, but ultimately, their integration with the grid operation will unlock their full potential.
The U.S. Department of Energy Solar Energy Technologies Office (SETO) has invested $20 million in two research programs to address these challenges. The first program, announced in 2012, sought to significantly improve forecasting accuracy and quantify the value of solar forecasts. Among its successes were NCAR’s development of WRF-Solar, a solar-targeted improvement of the benchmark Weather Research Forecast model, and IBM’s development of Watt-Sun, a multi-scale, multi-model, machine-learning based solar forecasting technology.
The second program, Solar Forecasting 2, which is now approaching the one year mark, will establish an industry framework to benchmark solar irradiance and power forecasts with clear rules and metrics that can be used by both forecast developers and end users. It also aims to improve forecasting accuracy in intra-day and day-ahead horizons, especially under partly cloudy skies. Finally, it seeks to develop new technologies that can integrate probabilistic forecasts with energy management systems so that solar power can help balance load with generation cost-effectively, in regulated or deregulated territories.
Early results of this program discuss probabilistic forecasts using large ensembles—many instances of the same model, just with different sets of parameters—and the performance of the benchmark WRF-Solar model under various cloud conditions. Equally important, the evaluation framework is being set up to connect with users via the Solar Forecast Arbiter portal, which also provides progress updates to an active stakeholder community. The desired outcome is that these results lead to improved forecasting that, along with reduced hardware costs, will allow large amounts of solar power to integrate cost-effectively into the grid.
Moreover, SETO is thinking about how to confront the challenges of the next decade, when the levelized cost of solar energy reaches $20 to $30 per megawatt-hour for large regions of the United States and the cost of battery energy storage is significantly reduced. The latter development will have dual effects, which are only partially aligned: substation-scale and, potentially, neighborhood- and home-scale storage will become much more affordable, and the electrification of the transportation sector will rapidly accelerate. While storage could be used to mitigate the uncertainty and variability of solar and wind power, the proliferation of electric vehicles brings a substantial electric load that could potentially be used as a resource, too.
These changes will be superimposed on an electric grid that interacts with its end users in new ways, if certain trends continue to find traction. When building managers and occupants can actively manage electricity consumption or supply services to local and regional grid operators, solar power forecasting becomes potentially more pervasive. Usually, only independent system operators need to know about impending output changes in utility-scale plants and high concentrations of rooftop photovoltaic (PV) systems. But that is changing: building managers, residential PV system aggregators, and even homeowners can benefit from knowing the next day’s forecast. Of course, if merchant PV power plants become more common on a grid where inverter-based generators are allowed to bid for a full range of grid services, their successful operation will depend on accurate forecasts through different time horizons.
While low-cost energy storage will help solar power become more ubiquitous and dispatchable, forecasting remains a cost-effective tool to integrate solar into the grid, since it doesn’t require capital expenditure for permits and equipment. What’s more, operating grid-connected batteries economically will require accurate forecasts of solar variability in order to select the charging strategy that takes maximum advantage of the available solar power and avoids undue cycling of the energy storage system. The same type of forecast can help with the optimal operation of flexible loads and other DER. When high penetration of variable renewables is combined with a high penetration of electric vehicles, multiday forecasts will become increasingly necessary to minimize the cost of vehicle charging and allow an efficient market for vehicle-to-grid services.
These plausible scenarios allow us to see how different aspects of the solar forecast have a two-way dependence on the penetration of certain technologies and business models. Forecasting of generation variability at sub-kilometer-level resolution may be necessitated by—and further empower—grid-transactive efficient buildings. Accurate multiday forecasts could be the basis for grid-transactive vehicles in a high-solar-penetration future. It is the responsibility of the stakeholders to identify the capability gaps and determine their urgency so that appropriate research and development efforts are prioritized and supported.
Tassos Golnas
Technology Manager
U.S. Department of Energy Solar Energy Technologies Office
Steve B says
Greetings, Tassos. I was speaking with someone about a related subject today. I am currently managing a dispatchable natural gas fired plant in California. We want to figure out how to understand when we have a high probability to run by looking at day ahead demand projections. Not working so well for us because we have no way to understand how many homes are producing power and thereby lowering the grid demand but not actually showing up in CAISO projections.