Quasi-Instantaneous Battery End-of-Discharge Time Prognosis with Non-Stationary Autoregressive Exogenous Inputs

Abstract

Reliable automation of engineering systems consider the implementation of condition monitoring routines to make decisions that can be carried out either manually, or through control loops. In recent years, great efforts have been made to incorporate prognostics into decision making, primarily through heuristic or data-driven approaches. However, being effective and efficient in this task is quite difficult. On one hand, it takes time to compute predictions, which makes prognostics-informed control unfeasible in real-time. On the other hand, until very recently, prognostics lacked a solid theoretical foundation to provide formalism and mathematical guarantees on prediction convergence. The concept of near-instantaneous prognosis was recently introduced, enabling prognostic analysis in nonlinear dynamical systems with maximal performance and stochastic convergence guarantees, under the assumption that future exogenous inputs follow strictly stationary stochastic processes. In this article, near-instantaneous prognosis is revisited to predict the End-of-Discharge time of batteries in which the discharge current, interpreted as exogenous input, is modeled as an autoregressive stochastic process. Furthermore, the strict stationarity condition is relaxed giving rise to a new variant presented as quasi-instantaneous prognosis, seeking to generalize this approach by removing this restrictive hypothesis in order to broaden its range of applications, especially in electrical engineering, such as in solar and wind resource prediction, or energy demand forecasting, to name some examples.