A Micromesh Multi-Scaled Features Extraction Network for Li-Ion Batteries SOH Estimation

The great progress and wide application of electronic products and electric vehicles have entailed more stringent requirements for the reliability of lithium-ion batteries. The State of Health (SOH) serves as a significant indicator for evaluating the condition of such batteries. However existing methods are lack of refined modeling for sequences and unable to make precise estimation for SOH. In this paper, a micromesh multi-scaled features extraction network (MMFEN) is proposed for accurately estimating SOH. A refined representation block is developed for heterogeneous elaborate feature extraction. Then, a multi-head convolution attention block is constructed to capture multi-scaled efficient state information. To demonstrate the superiority of MMFEN, experiments are conducted on the NASA and CALCE data published online and a real-world electric vehicles (EVs) data set which is collected from existing battery management systems. Comparing with traditional methods, MMFEN achieves remarkable performance with average root mean square error of 1.21% and mean absolute percentage error of 0.99% on NASA samples, 2.78% and 2.71% on CALCE dataset, while 2.39% and 2.08% on EVs data set, respectively.