Cell seletion through two-level basis pattern recognition with low/high frequency components decomposed by DWT-based MRA

Differences in discharging/charging current signal (DCCS) profiles with magnitude, time interval, abrupt changes result in erroneous information for selection of Li-Ion cells that have similar electrochemical characteristics when using the conventional pattern recognition of Hamming neural network (HNN). Thus, this work gives insight to the implementation of Li-Ion cells selection by two-level basis pattern recognition with approximation A_n and detail D_n components decomposed by discrete wavelet transform (DWT)-based multi-resolution analysis (MRA) for stable operation of Li-Ion battery pack. In the first level, 10 current profiles with different discharging/charging sequences are applied to a Li-Ion cell for obtaining different DCCS patterns. Low/high frequency components (A₅ and D₅) decomposed by DWT-based MRA using Daubechies wavelet (dB) with 5 scale are applied as characteristic parameters in the HNN for pattern recognition of the unknown DCCS. Specifically, through simple rule, an arbitrary DCCS´s current scale is little adjusted for matching with current scale of 10 representative DCCSs. In the second level, 15 representative discharging/charging voltage signal (DCVS) patterns considering previously selected DCCS in the first level, are used to the HNN for pattern recognition of an arbitrary DCVS. By proposed work, it is possible for elaborate identification of the representative DCCS and DCVS that most closely match those of an arbitrary Li-Ion cell. Consequently, this approach enables us to provide essential information for parameters estimation such as state-of-charge (SOC) and state-of-health (SOH) in addition to stable configuration of battery pack.