TODAY’S STUDY: THE STATE OF WIND AND SOLAR FORECASTING
A Review of Variable Generation Forecasting in the West, July 2013 — March 2014
R. Widiss & K. Porter and Dr. Debra Lew & Dr. David Hurlbut, March 2014 (Exeter Associates and NREL)
This report is based on a series of interviews with 13 operating entities (OEs) in the Western Interconnection about their implementation of wind and solar forecasting, jointly referred to as variable generation (VG) forecasting. This piece updates a report issued by the National Renewable Energy Laboratory (NREL) in 2012; it also covers several additional topics including sub-hourly scheduling, grid operator training, and forecasting for distributed solar resources. As in the 2012 report, the OEs interviewed vary in size and character; the group includes independent system operators, balancing authorities, utilities, and other entities that rely on VG forecasting.
VG forecasting is widely considered to be a key means of integrating wind and solar power efficiently and reliably as these resources become increasingly common. Indeed, in a recent report, grid operators from 18 countries identified wind forecasting as “the most important prerequisite for successfully integrating wind energy into power systems” (Jones 2011, p. xxiv).
VG forecasting remains a relatively new phenomenon in the West. Ten of the 13 OEs interviewed for this year’s report began using VG forecasts in 2007 or later. Each currently uses a wind forecast. In anticipation of rapid growth in solar generation, five OEs have recently begun working on in-house solar forecasts and two are utilizing outside sources. This report serves as a means for these companies to compare VG forecasting practices, lessons learned, and priorities with one another, as well as to share their experiences with state and federal regulators, market participants, national laboratories, and non-governmental organizations.
Costs and Benefits – The costs of wind forecasts have dropped dramatically since the 2012 report. This decline coincides with a shift toward testing or utilizing multiple vendors. Many of the OEs interviewed no longer view VG forecasting in a cost-benefit framework, regarding it instead as a necessity for maintaining electric reliability and scheduling resources effectively.
Cost Assignment – Only a few respondents partly or fully recover forecasting costs from variable generators. Many simply absorb the costs, possibly viewing them as relatively minor. However, the reportedly high cost of individual solar plant forecasts prompted at least one OE to turn to in-house forecasting.
Forecast Accuracy – Wind forecasting accuracy continues to improve incrementally. Participating OEs credit these gains to improved forecasting techniques and models, seasoned vendors, and growing portfolio size, all of which smooth the variability in VG output. Solar forecasting is at an early stage in the West, but at least one company is beginning to track solar forecasting accuracy.
Forecasting Uses – Nearly all interviewees use their wind forecasts for day-ahead unit commitment—a striking change since the 2012 report. This was consistent despite the entities’ diversity in size, proportion of renewables, and average monthly load. Intra-day unit commitment and reserves planning are the next most common uses, followed by a diverse array of uses often unique to a given entity.
Data Collection – Participating OEs have made few expansions, if any, to the types of meteorological data (wind speed, direction, temperature, pressure, humidity) and turbine status data they require of wind generators. However, two OEs have recently taken steps to increase the speed of data transmission from their generators, and reported that this change has greatly enhanced the value of their wind forecasts. Because solar forecasting is at an early stage, only a small number of responding OEs have solar data requirements in place.
Curtailments and Outages – Most interviewees factor turbine availability and/or outages into their forecasts so that they represent what generators are capable of producing, even if VG output is curtailed. Less than half of the OEs describe using curtailment information after the fact for calibrating forecast models and calculating performance metrics. Probabilistic Forecasting – Participants report that both ensemble forecasts and confidence intervals (CIs) are commonly used to address forecasting uncertainty. Yet many system operators reportedly ignore the CIs provided to them, choosing instead to use a single likeliest production value.
Distributed Solar Production – Distributed generation (DG) is commonly “invisible” to system operators, particularly for behind-the-meter resources connected at customer sites, which are netted out with the customer load. These resources cannot usually receive dispatch commands. Six of the OEs interviewed view the development of methods to forecast distributed solar production as an imminent need, and two see it as an eventual need. No consensus on how to forecast distributed solar generation has emerged.
Control Room Integration – Displays of VG forecasts in OE control rooms are nearly universal. Typically, these are automated feeds, sometimes provided by third-party forecasters. These displays are often accompanied by real-time weather or real-time generation data. Half the organizations interviewed are integrating forecast values directly into operations tools such as an Energy Management System (EMS).
Staff Familiarity – Though formal training is rare, staff members often coach their colleagues on an as-needed basis. System operators have developed a sense of familiarity with VG forecasts at most of the organizations interviewed. Four OEs also employ meteorologists to aid in interpreting VG forecasts.
Advice and Lessons Learned – Respondents’ advice for other utilities includes starting sooner rather than later as it can take time to plan, prepare, and train a forecast; setting realistic expectations; using multiple forecasts; and incorporating several performance metrics. Potential Regional Initiatives - Several of the OEs interviewed are against the creation of formal standards or guidelines for forecasting, suggesting that these would stifle innovation and impose “one-size-fits-all” methods upon unique situations. Others suggested that guidelines for data collection or a guideline determining resource adequacy for reserves would be helpful. A small number of interviewees advocated for further research and development (R&D) investments in forecasting.
Forecast Sharing - OEs were also split on the idea of sharing forecasts with other OEs. Some suggest that sharing forecasts and data would help improve VG forecasting. Others contend that sharing forecasts will not have much value unless reserves can be traded through such mechanisms as Energy Imbalance Markets (EIMs). Still others view VG forecasts as a source of competitive advantage for recipients and would oppose sharing them.
Sub-Hourly Dispatch - The changes documented since the 2012 report have been remarkable. Yet, it is also worth noting one practice that has not changed. Outside of the West, regional transmission organizations (RTOs) are now dispatching wind in five-minute markets as opposed to hourly schedules in the West, except for the Alberta Electric System Operator and the California Independent System Operator. The RTOs outside the West use equally fast forecast updates, taking advantage of the fact that forecasts are more accurate in short-term increments. Industry initiatives such as the EIM encompassing the California Independent System Operator, Nevada Power and PacifiCorp, as well as regulatory initiatives such as Federal Energy Regulatory Commission Order No. 764, may accelerate the adoption of this practice in the West.