State of power prediction joint fisher optimal segmentation and PO-BP neural network for a parallel battery pack considering cell inconsistency

Accurate state of power (SOP) of battery is critical for efficient control and stable operation of electric vehicles. Due to cell inconsistency and even varying degrees of discrepancy, especially, for the cells with large inconsistencies, accurately predicting SOP for a parallel battery pack faces challenges. To resolve these problems, a SOP prediction method joint Fisher optimal segmentation (FOS) and parrot optimizer-back propagation (PO-BP) neural network is developed. First, a quantitative differentiation strategy of the cell inconsistency based on weighted cosine similarity (WCS) is presented, which can describe the variation of cell similarity coefficient with maximum current deviation in parallel branch. Secondly, according to similarity coefficient and current deviation of the cells, a determination method using FOS for the degree of cell inconsistency is developed to identify cells with large inconsistencies. Finally, a power corrector based on the PO-BP neural network is developed to compensate for the SOP differences caused by the cells with large inconsistencies, thereby improving the SOP prediction accuracy. The experimental results verify the effectiveness of the developed method under various dynamic conditions, with a decrease of approximately 60 % in both the mean absolute error and the root mean square error of the SOP for a parallel battery pack.