Abstract
Learning the deterioration of a battery from charge and discharge data is associated with different non-random uncertainties. A specific methodology is developed, capable of integrating expert knowledge about the problem and of handling the epistemic uncertainty associated with conflicts in the available information. It is shown that the simple concatenation of charge and discharge data in a single training set leads to a biased model. Weak supervision techniques are used to assess the relative importance of subsets of the training data in the empirical loss function.
Type
Publication
In International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems
