Keywords: energy systems modelling, endogenous technical change, experience curves, uncertainty, stochastic programming
Introducing uncertain learning in an energy system model: a pilot study using GENIE
Energy system models based on experience curves are superior to conventional models in their treatment of the dynamics of technological development. However, assuming perfect foresight means that future learning rates are known with certainty, which is not realistic. An optimising model for the global electricity system, GENIE, has been extended to include imperfect foresight of learning rates. Technology cost trajectories are still determined by experience curves, but progress along the curves may follow alternative branches. Information about exactly which branch is currently being followed is not initially available, but may be subsequently revealed and acted upon. Unlike most applications of stochastic programming with recourse, the learning rate uncertainties are not resolved at a predetermined point in time. Instead, this information is only revealed once a certain threshold level of experience has been obtained for the particular technology. To minimise computational difficulty, only two technologies feature experience curve uncertainty - photovoltaic solar cells and fuel cells. The learning rates for these technologies can independently assume high or low values. Model results emphasise the importance of early learning investments in emerging energy technologies. The optimal hedging strategy calculated by GENIE involves significant early investments in both solar PV and fuel cells. An early commitment to emerging technologies is not only a good investment plan when high learning rates are expected, but also an efficient hedging strategy when future learning rates are uncertain. A sensitivity analysis also shows that this investment strategy is surprisingly robust even if high future learning rates are regarded as improbable.