Uncertainty-Aware Data Selection Framework for Parameter Estimation with Application to Li-ion Battery

Abstract

Data selection is a technique for improving parameter estimation accuracy by strategically selecting data segments from a measured data set as the inputs to an estimation algorithm. Traditional selection criteria have been heuristic or sensitivity-based, without consideration of uncertainties in measurement, model, or parameter. In this paper, we propose an uncertainty-aware data selection framework that selects data segments based on the capacity for the ingrained data structures to mitigate the influence of system uncertainties on the estimation result. The framework comprises two components: the data quality rating, which is a metric for evaluating the uncertainty-propagating data structures of a data segment; and the data selection algorithm, which automatically integrates the data selection and estimation procedures. A demonstration of the framework is presented in a Li-ion battery application, where two electrochemical parameters are separately estimated under random input data sets. The data selection framework reduced estimation errors by up to two orders of magnitude when compared with the conventional approach of estimating without data selection.