Variability is not a new issue for grid operators, as they are accustomed to loads varying on similar time scales and the fact that sometimes power plants and transmission infrastructure fail.  These “traditional” sources of variability are well understood and significant effort has been devoted to developing systems to manage these sources of variability. 

Until PV reaches a certain level of penetration on the grid, it is essentially noise in the system with no noticeable impacts on reliability.  At some point, however, PV penetration levels results in this new, less familiar and more random form of variability impacting grid operations.  This is already happening in certain service locations such as several of the Hawaii islands where installed PV represents a significant portion of the overall generation.  Furthermore, some locations in California, New Jersey, and Arizona have reached threshold levels where PV variability is beginning to have noticeable impacts on grid operations.

Solar forecasting should be the first response to managing the variable nature of solar energy production, before the more costly strategies of energy storage and demand response systems are put in place.  Furthermore, once a forecasting system is in place, it provides additional benefits through the optimized use of these demand-side resources.

Sophisticated tools have been developed using data from weather models and cloud images from geostationary satellites or total sky imaging devices to provide solar power forecasting services for various time horizons, from day-ahead forecasts of power production from PV systems to sub-hourly forecasts.   Modeling the random nature of clouds, which drives the short-term variability in PV system output, is an active area of research and will likely lead to improved accuracies of sub-hourly forecasting. 

Forecast can be produced for individual solar arrays or for numerous arrays within a defined geographic area.  The geographic dispersion of the installed PV capacity for area forecasts reduces error due to the fact that localized weather conditions do not impact the installed capacity uniformly.  Tom Hoff of Clean Power Research and Richard Perez of SUNY Albany’s Atmospheric Sciences Research Center demonstrate this phenomenon empirically using correlation coefficients between geographically dispersed PV systems, which decline predictably as a function of distance.  Increasingly, grid operators are interested in ramp forecasts, which seek to predict significant changes in solar output within a predefined time step (e.g. 15-minutes).

Grid operators have been using wind forecasts for close to a decade to manage the variability of wind farm output.  The California Independent System Operator (CAISO) has been using a wind forecasting service since 2004, and all the other major ISOs/RTOs (regional transmission organizations) currently utilize central wind forecasting services for reliability planning and market operations.  CAISO has begun to experiment in recent years with integrating solar forecasting into planning and market operations.  The CAISO solar forecast is provided by AWS Truepower, a leading renewable energy project development and operations solutions provider.

Market reforms in several wholesale power markets now allow intermittent resources to participate in wholesale power markets on par with conventional sources of generation.  Thus some wholesale power markets allow wind forecasts to be used to bid wind energy into day-ahead, hour-ahead, and real-time markets.  Renewable energy forecasting has a variety of other uses, including developing bidding strategies for owners of renewable energy plants to system reliability planning. 

While there is much we can learn from wind power forecasting and centralized solar system can benefit from recent market reforms, there are unique challenges for solar given the largely distributed nature of solar deployment.  Behind the meter solar is largely invisible to grid operators and the variability becomes manifest in changes in the net load to be served.  Thus, solar forecasting can provide value by feeding into existing load forecasts models to improve net load forecasts.

Solar forecasting’s accuracy is enhanced when historic production data is integrated into the forecast model to “train” the models.  Modeled historic production data using satellite cloud cover images can be used as a substitute for actual historic output; this requires access to central database of technical specification and array characteristics for thousands of installations.  Clean Power Research, a solar forecasting provider, is working with the CAISO to build systems specifically for forecasting the tens of thousands of distributed PV systems across their service territory.  The CAISO load forecast saw a statistically significant increase in accuracy when driven with behind the meter production inputs from Clean Power Research’s forecasting tool, SolarAnywhere FleetView.

In pursuit of policies to encourage solar energy deployment, policymakers and regulators should consider standards that require grid-connected solar plants to monitor and report real-time production data to enhance forecasting accuracy over time and establish the requirement that information on behind the meter solar including array orientation, angle, shading, etc. be made available to the forecasting community.

The solar age is upon us; while the variable nature of solar energy production is real, we have tools that will allow us to efficiently manage this variability.  Eventually, energy storage and demand response will allow even greater levels of solar PV generation.  Solar forecasting, however, represents the lowest-cost, near-term opportunity to manage the variable nature of solar energy production.  Today there are over a dozen companies globally that provide solar forecasting services, the low-hanging fruit of solar integration.  Let’s take a bite!

Lead image: Fruit trees via Shutterstock