Big data driven lithium-ion battery modeling method based on SDAE-ELM algorithm and data pre-processing technology

Abstract As one of the bottleneck technologies of electric vehicles (EVs), the battery hosts complex and hardly observable internal chemical reactions. Therefore, a precise mathematical model is crucial for the battery management system (BMS) to ensure the secure and stable operation of the battery in a multi-variable environment. First, a Cloud-based BMS (C-BMS) is established based on a database containing complete battery status information. Next, a data cleaning method based on machine learning is applied to the big data of batteries. Meanwhile, to improve the model stability under dynamic conditions, an F-divergence-based data distribution quality assessment method and a sampling-based data preprocess method is designed. Then, a lithium-ion battery temperature-dependent model is built based on Stacked Denoising Autoencoders- Extreme Learning Machine (SDAE-ELM) algorithm, and a new training method combined with data preprocessing is also proposed to improve the model accuracy. Finally, to improve reliability, a conjunction working mode between the C-BMS and the BMS in vehicles (V-BMS) is also proposed, providing as an applied case of the model. Using the battery data extracted from electric buses, the effectiveness and accuracy of the model are validated. The error of the estimated battery terminal voltage is within 2%, and the error of the estimated State of Charge (SoC) is within 3%.