Experience curve models in technology cost forecasting
: The case of solar Photovoltaic modules

  • Bdour Sbeih

Student thesis: Doctoral Thesis

Abstract

Energy technology models are required to enable strategic plans for decarbonisation. This requires accurate modelling of cost reductions due to technological learning. The premise of this research is to determine if new experience curve models could be implemented to reduce the error in cost estimates for solar PV technologies.

Experience curves are used in technology cost models, where technology costs decline as experience is gained through production and implementation. Since Wright’s observation of the phenomenon in 1936, experience curves have been conventionally written as a linear function of cost and production, assuming constant learning over time.

This research investigates the constant learning rate by evaluating the experience curve slope in relation to the shape of the model. It compares conventional (linear) and contemporary (nonlinear) experience curve functional forms to determine the most accurate model.

The application of nonlinear experience curve models, that mathematically allow for a flattening effect, is not well explored in literature on emerging technologies in general, and energy technologies to our knowledge. Simplicity and ease of use are among reasons of the popularity of conventional models.

The purpose of this research is to investigate the reliability of contemporary experience curve models in forecasting technological cost compared to Wright’s conventional model. This analysis specifically examines whether the implementation of Gompertz and the Logistic nonlinear models would reduce the error in cost estimates in comparison to Wright’s power-law curve. It is a detailed theoretical and statistical review on the performance of these models in the analysis of technological learning. The statistical comparison is performed using global Solar Photovoltaic (PV) modules production data.

By conducting a regression analysis, the results showed a statistically significant reduction in error in nonlinear models through the measurement the two error terms, Sum of Squared Errors and Mean Absolute Percent Error. This thesis explains in detail how testing was conducted to compare the different experience curve methodologies, using 25 years production data for solar PV modules cumulative installed capacity and inflation-adjusted costs. The research further justifies the theoretical necessity for models that explain the diminishing technological learning rates. It is acknowledged that, in addition to technological progress, addressing global challenges through innovation also involves social, political and economic changes.

Date of AwardFeb 2023
Original languageEnglish
Awarding Institution
  • University of Brighton
SupervisorRob Hayward (Supervisor) & Timothy Laing (Supervisor)

Cite this

'