Paper Title
A Deep Neural Network Structure for Li-Ion Battery State of Charge Estimation and Hardware Implementation in Electric Vehicles

Abstract
This study proposes a battery model that predicts the Li-ion SOC of electric vehicles using a deep neural network (DNN) structure. Unlike most machine learning-based models, which employ large neural networks with billions of parameters to determine target functions, this study aims to achieve high accuracy rates using smaller DNN models. The study utilizes data collected from the Center for Advanced Life Cycle Engineering (CALCE) and verifies the accuracy of the DNN model using Federal Urban Driving Schedule (FUDS) drive cycle datasets. The study also uses actual electric vehicle battery data from Argonne National Laboratory for testing the DNN model. This approach makes the hardware implementation of deep neural network estimation more efficient. The study develops and evaluates a series of DNN models with varying numbers of neurons and finds that increasing the number of neurons does not reduce the error rate, but it increases costs. Additionally, the study reduces the complexity of the neural network algorithm circuit using battery measurement circuit accuracy analysis[2]. The objectives of this study are to demonstrate that DNN models can accurately estimate the SOC of Li-ion batteries with sufficient battery data, identify the optimal number of neurons for SOC estimation, and implement the testing part of the DNN model in hardware. Keywords - Battery, Dnn, State Of Charge