Revisiting Learning for Utility-Scale Wind and Solar in the United States

Abstract

SummaryLearning curves play a central role in power sector planning. One challenge is to address welldocumented limitations in the accuracy of learning-based projections. We improve upon past learning curves for utility-scale wind and solar in three ways. First, we generate plant-level estimates of the levelized cost of energy (LCOE) in the United States, and then use LCOE, rather than capital costs, as the dependent variable. Second, we normalize LCOE to control for exogenous influences unrelated to learning. Third, we use segmented regression to identify change points in LCOE learning. We find fullperiod learning rates of 15% for wind and 24% for solar, with accelerated learning of 40%-45% in recent years. We conclude that (normalized) LCOE-based learning provides a more complete view of technology advancement than much of the existing literature affords. Models that do not account for recent accelerated LCOE reductions or endogenous LCOE-based learning may underestimate future cost reduction.