The automotive industry is currently investing heavily to pivot from conventional internal combustion engines to electric powertrains. The battery pack still makes up a large portion of each electric vehicle’s cost, making management and control of the battery crucial for this transition. With improved state of charge estimation an increased amount of energy can safely be extracted from each battery pack, therefore increasing the range, or allowing smaller packs to be utilized. For this study, traditional equivalent circuit models and machine learning methods are compared for battery state of charge estimation across a temperature range of -10°C to 25°C. The machine learning models explored include support vector regression, feedforward neural networks and recurrent neural networks. The dataset used was derived from various tests and drive cycles performed on a Panasonic 18650PF NMC Cell in a temperature chamber. Based upon the results, it is evident that equivalent circuit models perform very well at higher temperatures, but struggle to capture the highly nonlinear characteristics of batteries at lower temperatures. On the other hand, when properly trained and parameterized, machine learning models can be much more effective at capturing the battery’s characteristics at low temperatures while maintaining their computational requirements low.
Comparing Traditional and Machine Learning Models for Battery SOC Calculation
2022-06-15
442895 byte
Conference paper
Electronic Resource
English
Comparing discrete choice and machine learning models in predicting destination choice
DOAJ | 2024
|Comparing discrete choice and machine learning models in predicting destination choice
Springer Verlag | 2024
|British Library Conference Proceedings | 2003
|Flexible oneanchor retaining building models calculation results comparing with experimental data
DOAJ | 2019
|