A Combined DNN-NBEATS Architecture for State of Charge Estimation of Lithium-Ion Batteries in Electric Vehicles
M Kannan, K. Sundareswaran, P. Srinivasa Rao Nayak, Sishaj P. Simon
Abstract
In this paper, a new hybrid architecture combining Deep Neural Network (DNN) and Neural Basis Expansion Analysis for Time Series (N-BEATS) is proposed first time for estimating the State of Charge (SoC) of Lithium-ion batteries. The input and target vectors used for training and testing the proposed hybrid architecture is experimentally determined using a low-priced microcontroller. The proposed hybrid architecture is trained under different operating conditions, tested with untrained data, and is shown capable of accurately estimating the SoC. The architecture is also trained and tested with four different dynamic loading profiles. Further, existing and well-established neural networks, namely Long short-term memory (LSTM), Bidirectional long short-term memory (Bi-LSTM), and Gated recurrent unit (GRU), are employed and tested under identical conditions, and results are compared. Extensive computation results are presented for various temperatures and battery chemistries, and comparison highlights the superior performance of the newly proposed hybrid architecture.