Hybrid Single-Particle Model with Thermal Dynamics and a Neural Network to Predict Lithium-ion Battery Voltage

In the context of the energy transition, lithium-ion batteries are essential for many applications, ranging from consumer electronics to electric vehicles. To fully exploit their potential, it is crucial to have reliable models for estimating, monitoring, and diagnosing their condition. Our study proposes an innovative approach that couples the single-particle model with thermal dynamics (SPM _ T) and a multilayer perceptron (MLP) to improve the estimation of the terminal voltage of batteries. Simulation results show that our method increases the accuracy of voltage estimation by up to 80% depending on the C-rate, compared to the SPMT alone, while maintaining computational simplicity. The developed hybrid model also presents flexibility for various cell chemistries and maintains high accuracy in the face of aging. Moreover, the learning process provides efficient generalization with a limited number of scenarios. The results obtained in this study will serve to design more reliable battery management systems (BMS).