State-of-charge estimation of Li-ion batteries using Deep Neural Networks: A machine learning approach
Lithium-ion batteries have many advantages when compared to other batteries such as high cycle life and low self-discharge, due to which they are heavily used in portable electronics, Electric Vehicles (EV), smart-grid technology and unmanned aerial vehicles. EV powered by Lithium-ion batteries are particularly advantageous to mitigate air pollution created by petrol and diesel powered vehicles. In 2015, 50% (53 million tons) of all nitrogen oxide air pollutants in the world was attributed to transportation sector. Some countries such as Norway are aiming to ban petrol and diesel powered vehicles by 2025 and thus usage of EV will be more significant.
An accurate estimation of the residual charge (State of Charge a.k.a. SOC) in battery is essential for safe and reliable operation of EV. The two main traditional methods for estimation of SOC such as Open Circuit Voltage based techniques and Coulomb counting are replaced by more sophisticated methods like Kalman filter and observer based methods. These complex methods require hand engineering of models that approximate the battery behavior.
Modern research works show application of different Machine Learning (ML) algorithms in combination with Kalman filter to achieve an acceptable error rate in SOC estimation. The most recent one describes an end-to-end system based on Deep Feedforward Neural Network (DNN). The optimized model has 4 layers, 56 hidden units, 4 inputs (voltage, temperature, average current and average voltage of battery) and SOC as output, with lowest Mean Absolute Error of 0.61%, which is better than other models published in the past. DNNs can map observable signals from battery like voltage, current and temperature to directly estimate SOC without additional components, providing reliable SOC estimation under different ambient temperatures (-20 to +25 oC) with a single DNN, reducing computation time and ease the modeling process by avoiding hand-engineering and saving time/efforts.
Reference: Chemali, E., Kollmeyer, P. J., Preindl, M., & Emadi, A. (2018). State-of-charge estimation of Li-ion batteries using deep neural networks: A machine learning approach. Journal of Power Sources, 400, 242-255.