Voltage Prediction in Lithium-Ion Batteries: Integrating Physics-based Model with Neural Networks for Enhanced Accuracy

Lithium-ion battery technology, a cornerstone of modern energy storage solutions, has been extensively researched, particularly in the realm of mathematical modeling. However, challenges persist in balancing model accuracy with computational efficiency. Addressing these challenges, this study introduces a novel hybrid model that fuses principles of physics-based modeling with the computational processes of the FNN. The core electrochemical framework of the model is based on the well-regarded Doyle Fuller Newman (DFN) model, which effectively captures the internal states of the battery. These states are then utilized as key inputs to train the neural network. This innovative hybrid approach not only enhances accuracy but also significantly reduces complexity. Its versatility is demonstrated through its capacity to accurately predict battery voltage across various C-rates and a broad spectrum of temperatures, including extreme conditions. Additionally, the model’s performance was rigorously tested under varying current loads, showcasing its efficacy and potential applicability in diverse real-world scenarios.