The increasing adoption of electric vehicles (EVs), efficient and accurate battery modeling has become crucial for reliable performance evaluation and control system design. However, maintaining high accuracy in simulations generally requires complex computations, which can limit real-time applicability and scalability. High-fidelity battery models often require significant computational time, making them unsuitable for real-time simulations and large-scale system integration. This paper presents the application of Simulink Reduced Order Models (ROM) to simplify the simulation of EV batteries while maintaining acceptable levels of accuracy.
The EV simulation environment has been developed in MATLAB/Simulink to analyze Battery Management System (BMS) control system design and assess EV system level performance. This simulation platform consists of BMS and other important EV controller models and high-fidelity plant models for battery and powertrain systems. While these high-fidelity models enable accurate virtual testing and control logic development, they also impose substantial computational requirements, leading to slower simulation performance. This paper addresses these challenges by proposing a Reduced-Order Modeling (ROM) approach, leveraging Artificial Intelligence (AI) techniques to significantly improve system-level simulation efficiency.
In this study, a computationally intensive high-fidelity EV battery pack plant model, which was originally modelled using Simscape battery library was replaced with trained Neural State Space (NSS) ROM model using MATLAB/Simulink tool. A low-order nonlinear ROM based on the Neural State Space (NSS) architecture is developed using deep learning methods, effectively acting as a surrogate for the computationally intensive high-fidelity battery model. The trained ROM was integrated into the Simulink system-level simulation platform and benchmarked against the original high-fidelity model. Simulation results demonstrate that the ROM effectively captures the essential dynamic behavior of the battery while significantly reducing computational costs compared to the baseline high-fidelity Simscape model.
The proposed approach achieves a notable reduction in simulation time while maintaining acceptable accuracy under the evaluated drive cycle and operating conditions. This work demonstrates the practicality of ROM-based modeling as a key facilitator for efficient EV battery analysis, design optimization, and control strategy development.