Physics-based learning for advanced battery modeling

Lecture Abstract

Integrating physics-based and data-driven approaches has emerged as a promising strategy for modelling dynamic systems. Yet, existing works primarily focus on using data-driven surrogates as replacements for physics-based models, often sacrificing accuracy for computational speed and losing critical elements like generalisability, adaptability, and interpretability that are essential for control applications. This lecture introduces a novel framework for physics-based machine learning, termed Model-Integrated Neural Networks (MINN), designed to learn reduced-order models of the complex physics of systems governed by partial differential-algebraic equations. MINN provides a systematic approach to develop optimally simplified models that balance physical insight, numerical accuracy, and computational efficiency.

We apply the proposed neural network architecture to model the electrochemical dynamics of lithium-ion batteries and show that MINN is extremely data-efficient to train while being sufficiently generalizable to previously unseen input data, owing to its underlying physical invariants. The MINN battery model has an accuracy comparable to the first principle-based model in predicting both the system outputs and any locally distributed electrochemical behaviors but achieves two orders of magnitude reduction in the solution time. Unlike physics-informed neural networks, which only approximate solutions to autonomous systems, MINN combines neural networks and physics-based models directly within the network architecture. This integration enables efficient training and can be extended to a range of control-oriented modelling scenarios.