Huzaifa Rauf a b c d, Muhammad Khalid e f g, and Naveed Arshad c d
a Department of Electrical Engineering, Lahore University of Management Sciences (LUMS), Sector U, Phase 5 D.H.A, Lahore, 54000, Pakistan
b Center for Advanced Life Cycle Engineering, University of Maryland, College Park, MD, 54000, United States
c Department of Computer Science, Lahore University of Management Sciences (LUMS), Sector U, Phase 5 D.H.A, Lahore, 54000, Pakistan
d LUMS Energy Institute, Sector U, Phase 5 D.H.A, Lahore, 54000, Pakistan
e Electrical Engineering Department, King Fahd University of Petroleum and Minerals (KFUPM), Dhahran, 31261, Saudi Arabia
f Interdisciplinary Research Center for Renewable Energy and Power Systems (IRC-REPS), KFUPM, Dhahran, 31261, Saudi Arabia
For more information about this article and related research, please contact Prof. Michael G. Pecht.
Abstract:
Lithium-ion batteries are a key storage technology for electric vehicles and renewable energy applications. However, the complex degrading behaviour of batteries impacts their capacity and lifetime. Thus, battery capacity loss prediction is crucial for ensuring the longevity, safety, and reliable operation of the battery. This research proposes a smart feature selection (SFS) strategy-based machine learning framework for battery calendar and cyclic loss prediction. The presented methodology selects input parameters from the battery data of the current time step as well as the previous time step which are then utilized for model training and testing. Results demonstrate that the proposed SFS method in combination with the ML algorithms enhances the prediction accuracy and reduces the mean absolute error for all the machine learning algorithms applied in this study. The proposed SFS method is capable of excavating the useful features, therefore offering good generalization ability and accurate prediction results for capacity loss of the lithium-ion battery under real EV usage conditions. Furthermore, the results also depict that the performance accuracy of ML methods for battery calendar and cyclic loss prediction improves when combined with the SFS method. Greater improvement in prediction accuracy of battery capacity loss is observed for Gaussian Process Regression (GPR), random forest (RF), and XGBoost methods when applied in combination with the proposed SFS. This is the first-known feature selection-based ML application that is utilized to independently perform battery calendar and cyclic loss prognosis.
This article is available online here and to CALCE Consortium Members for personal review.