Hybrid Physics Informed Neural Network for Thermal Aware Electric Vehicle Battery Management

Abstract

One of the most important issues with electric vehicle battery systems is thermal management because too high or too low temperature changes may impair battery performance, shorten lifespan, and become safety hazards. This paper suggests a Hybrid Physics Informed Neural Network (PINN) design of managing electric vehicle batteries thermally. The given model combines both physics-driven thermodynamic modeling and deep learning to provide proper and physically realistic battery temperature forecasting. As input features, battery operational parameters, including current, voltage, state of charge, and ambient temperature are used, whereas thermal constraints are included in the neural network training process. The hybrid model acquires nonlinear thermal dynamics and adheres to thermodynamics. Results of experimental analysis based on real-world battery data show the best performance under comparison to thermal models and purely data-driven analysis. The accuracy of the proposed model is 99.21%, Mean Absolute error is 0.44°Cand root mean square error is $\mathbf{0. 6 1}^{\boldsymbol{\circ}} \mathbf{C}$. The model also has real-time prediction with response time of less than 10 ms, which makes it an appropriate model to be deployed practically in battery management systems. Due to the integration of the physical constraints, predictive stability and predictive capabilities is enhanced. The suggested Hybrid PINN architecture offers an effective and smart methodology in achieving better safety, performance and efficiency of battery in electric vehicles.