Abstract
Mathematical modeling of lithium-ion batteries (LiBs) is a primary challengein advanced battery management. This paper proposes two new frameworks tointegrate physics-based models with machine learning to achieve high-precisionmodeling for LiBs. The frameworks are characterized by informing the machinelearning model of the state information of the physical model, enabling a deepintegration between physics and machine learning. Based on the frameworks, aseries of hybrid models are constructed, through combining an electrochemicalmodel and an equivalent circuit model, respectively, with a feedforward neuralnetwork. The hybrid models are relatively parsimonious in structure and canprovide considerable voltage predictive accuracy under a broad range ofC-rates, as shown by extensive simulations and experiments. The study furtherexpands to conduct aging-aware hybrid modeling, leading to the design of ahybrid model conscious of the state-of-health to make prediction. Theexperiments show that the model has high voltage predictive accuracy throughouta LiB's cycle life.