Chained Gaussian processes with derivative information to forecast battery health degradation

In order to estimate the value of a battery design, a key objective of manufacturers is to model battery health degradation. This estimation should be paired with a precise uncertainty modeling to quantify associated financial risks. However, experimental data are often limited in practice by their cost and the necessary time to perform a complete test. Prolonging previous work on uncertainty quantification, we tackle the issue of reducing testing time. To do so we propose to improve the forecasting ability of our model by integrating the available prior knowledge on derivatives. This is done through an extension of the chained Gaussian process framework is which is introduced in this paper.