An Effective Deep Neural Network Method for Prediction of Battery State at Cell and Module Level. Issue 7 (17th April 2021)

• Record Type: Journal Article
• Authors: Nguyen-Thoi, T.; Cui, Xujian; Garg, Akhil; Gao, Liang; Truong, Tam T.
• Is Part Of: Energy technology. Volume 9:Issue 7(2021)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2021-04-17
• Subjects: battery modules -- deep neural network -- single-cell batteries -- state of charge; Energy development -- Periodicals; Power resources -- Periodicals; 333.79
• DOI: 10.1002/ente.202100048
• Languages: English
• ISSNs: 2194-4288

Abstract

Abstract : The fast and accurate prediction of the battery state used in electric vehicles plays a very important role to guarantee the safe and reliable operation of the battery during its service life. Therefore, this study proposes an effective deep neural network (DNN) method for predicting the state of charge (SOC) of the single‐cell battery and the priority of the discharge of the battery module. In the proposed method, DNN models are constructed, trained, and evaluated based on experimental datasets. The mini‐batch and dropout techniques are applied to increase the training rate and eliminate the overfitting phenomena during the training process, respectively. In addition, various optimizers and activation functions are investigated. Hyperparameters of the neural network are inspected to determine the optimal network architecture. The performance and applicability of the DNN are illustrated through two different examples. In the first example, the priority for the discharge of each of the six battery modules is predicted by the DNN. Meanwhile, the DNN is used to evaluate the charge and discharge capacities of the single‐cell battery in the second example. The predicting results of the DNN are compared with those in the literature and experimental results to demonstrate the reliability of the proposed method. Abstract : Herein, a multi‐input and multi‐output deep neural network (DNN) method is successfully developed, based on experimental data for the evaluation of states of batteries utilized in real‐world electric vehicles. The obtained results show that the DNN is superior to the traditional artificial neural network (ANN) and is an effective and highly accurate method for predicting battery states.