The Electric Power Research Institute (EPRI) developed the U.S. Regional Economy, Greenhouse Gas, and Energy (US-REGEN) model, under the PRISM 2.0 Project. This model can assess the impact of various climate, energy, and environmental policies on the electric power sector, the energy system, and the overall U.S. economy. This report compares the technology learning rates implied by the exogenous cost specifications used in the US-REGEN model to those found in a literature review of observed learning curves.
Understanding how the performance and cost of energy supply technologies changes over time is of key importance for analysts and decision-makers concerned with large-scale energy systems and their impacts on the economy and the environment. Studies over the past several decades have documented how the costs of energy technologies have evolved in the past. Based on historical data, researchers have proposed various types of “learning curves” (or experience curves) to relate the future cost of a technology to key parameters such as installed capacity, R&D spending levels, and other factors such as economies of scale, market structure, materials cost, and efficiency improvement.
The most widely used model of learning estimates the price or performance of a technology as a log-linear function of its cumulative installed capacity. While suitable for some situations, this model form provides an overly simplistic characterization of technological change, which in reality is driven by complex dynamic relationships involving multiple factors. Just what those factors are, however, and how they influence technology change, remains a subject of ongoing research. Furthermore, while the simple log-linear relationship between cost and installed capacity has been widely studied for some technologies such as wind and solar, this approach is ill-suited to projections for new technologies that are just gaining market share or have yet to be scaled up for commercial application.
This report provides a comprehensive review of the recent literature on learning curves applicable to the electricity supply technologies available in the US-REGEN model to gain an understanding of the best ways to represent technological change in the model.
The report includes (1) a comprehensive literature review of technology learning models applicable to electric power generation technologies; (2) characterization of the state of the art and key findings stemming from that literature review; (3) review and characterization of other large-scale computer models that incorporate technology learning; and (4) from that review, lessons and insights potentially applicable to the development and use of EPRI’s REGEN model.
The review found that there is a wide variation in reported learning rates. Some studies include both learning-by-doing and learning-by-researching (reflecting R&D spending) and report both values. In general, there are wide variations even within the same technologies and no clear trend of learning rates associated with a certain type of technology, time period, or region. Researchers found a narrower range of smaller learning rates associated with fossil power plants, whereas renewable technologies (wind, solar, biopower) have a wide range of learning rates including values as high as 45% to 53%. With the exception of nuclear power, all the studies reviewed report cost reductions with increased installed capacity.
There are two key categories of uncertainties associated with the application of experience curves. One is the learning curve itself; the other concerns the conclusions drawn from the use of learning curves. Because of these uncertainties, an expanded and more systematic use of sensitivity studies, especially the testing of alternative learning curve formulations, should be pursued to better and more fully characterize their implications for projected cost reductions and rates of technological change.
Application, Value, and Use
This report provides a comprehensive overview of current literature on learning curves and of the use of learning curves in a number of large-scale energy-economic models. The report will inform the reader of work that has already been conducted in the assessment of endogenous technology change and of key considerations for modeling technology improvements moving forward.