Just-in-time Point Prediction Using a Computationally-efficient Lebesgue-sampling-based Prognostic Method: Application to Battery End-of-discharge Prediction

Abstract

Battery energy systems are becoming increasingly popular in a variety of systems, such as electric vehicles. Accurate estimation of the total discharge of a battery is a key element for energy management. Problems such as path planning for drones or road choices in electric vehicles would benefit greatly knowing beforehand the end of discharge time. These tasks are generally performed online and require continuously quick estimations. We propose a novel prognostic method based on a combination of classic Riemann sampling (RS) and Lebesgue sampling (LS) applied to a discharge model of a battery. The method utilizes an early and inaccurate prediction using a RS-based method combined with a particle-filter based prognostic. Once a fault condition has been detected, subsequent Just-in-Time Point (JITP) estimations are updated using a novel LS-based method. The JITP prediction are triggered when the Kullback-Leibler divergence between the probability density functions (PDF) of the long-term-based prediction and the last filtered state reaches a threshold. The CPU time needed to execute a procedure is used as a measure of the computational resources. Results show that this combined approach is several orders of magnitude faster than the classical prognosis scheme. The combination of these two methods provides a robust JITP prognosis with less computational resources, a key factor to consider in real-time applications in embedded systems.