Cimen H.2024-12-102024-12-102024979-833153149-2https://doi.org/10.1109/IDAP64064.2024.10710833https://hdl.handle.net/20.500.13091/96728th International Artificial Intelligence and Data Processing Symposium, IDAP 2024 -- 21 September 2024 through 22 September 2024 -- Malatya -- 203423In recent advancements in energy storage systems, accurate state-of-charge (SOC) estimation for lithium-ion batteries is crucial for optimizing performance and longevity. This study investigates the impact of instance normalization on the generalization capability of convolutional neural networks (CNNs) used for SOC estimation. Instance normalization, a technique initially popularized in image style transfer, normalizes each feature map independently, potentially improving model robustness and adaptability to diverse battery conditions. We evaluate the performance of ResNet-18 model with and without instance normalization on a dataset comprising various operational conditions of Li-ion batteries. Our experiments demonstrate that integrating instance normalization into the ResNet-18 architecture enhances the model's ability to generalize across different battery states, leading to more accurate SOC predictions. This improvement is particularly significant in scenarios involving novel or previously unseen conditions, highlighting the potential of instance normalization to enhance the reliability of SOC estimation systems in real-world applications. © 2024 IEEE.eninfo:eu-repo/semantics/closedAccessconvolutional neural networkgeneralization capabilityinstance normalizationli-ion batterystate of chargeBattery management systemsBattery storageConvolutional neural networksLithium-ion batteriesBattery energy storageConvolutional neural networkGeneralization capabilityInstance normalizationIon batteriesLithium ionsNetwork-basedNormalisationState-of-charge estimationStates of chargesState of chargeConference Object10.1109/IDAP64064.2024.107108332-s2.0-85207899414