Work

Rationalizing machine learning models for real-time health prognosis of lithium-ion batteries

Year: 2025

Type: article

Abstract: Accurate personalized real-time health prognosis of Lithium-ion batteries is critical for the operational safety and reliability of energy storage systems. State of Health (SoH) and Remaining Useful Life (RUL) are two main tasks to be solved. Prevalent machine learning approaches often employ computationally intensive deep neural network models and assume a synergistic relationship between SoH tracking and RUL prognosis.This paper introduces a conceptual framework—based on four concepts to rationalize prognostic model design. We systematically deconstruct and improve a state-of-the-art deep transfer model. We propose several deep variants and a lightweight Echo State Network featuring a novel reservoir structure.We determine that distinct strategies optimized for their degradation dynamics are advantageous for RUL and SoH. Comprehensive experiments across three diverse battery datasets demonstrate how our models outperform deep transfer learning for RUL and SoH. Our best models achieve important RMSE improvements of 40% and 91%, for RUL and SoH, respectively, on the largest dataset.

Source: Array

DOI: 10.1016/j.array.2025.100577

Authors: Mario Mallea, Àngela Nebot, Francisco Mugica

Institution: Universitat Politècnica de Catalunya

Language: English

Topic: Advanced Battery Technologies Research

Subfield: Automotive Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Affordable and clean energy

Open Access status: gold

Estimated APC paid: $1,680