Maréchal, FrançoisLi, XiangSchnidrig, JonasSouttre, Matthieu2022-02-222022-02-222022-02-222022-02-22https://infoscience.epfl.ch/handle/20.500.14299/185637Energy technologies are subject to dramatic cost changes. Indeed, the recent trend in investment costs (expressed in [USD/kW]) for the main renewable energy technologies showed an important cost reduction (e.g. solar PV and wind) or increase (e.g. hydro-power, geothermal energy) during the last decade. Taking these cost variations into account within the scope of a prospective energy system model such as Energyscope is therefore very important. While data concerning well-known energy technologies like solar PV and wind is widely available, emerging technologies such as electrolysis and CCUS (carbon capture, utilization and storage) have very few reliable data. Predicting their future cost is therefore a challenging task. To this end, the learning curve theory has been used. It has been widely used to model the cost reduction achieved in the industry via the learning-by-doing process, and can easily be transposed to energy technologies. The assessment of carefully selected learning curve functions applied to energy technologies historical data results in important cost changes until 2050. Indeed, the cost reduction between 2020 and 2050 of the cost-decreasing technologies under study are between 36% (onshore wind) and 74\% (residential solar PV), with a mean over the studied technologies that equals 50%.Energy systemCost evolutionMachine LearningArtificial IntelligenceLearning curvesTechnologiesPVWindHydrostudent work::semester or other student projects