Vector processor for online lithium-ion battery capacity prediction

Battery capacity prediction in aerospace systems is a computationally expensive problem. In this paper, we propose a novel field programmable gate array-based (FPGA) vector processor to reduce latency in this application. This processor architecture is optimized for the kernel recursive least squares (KRLS) algorithm, and used to perform online regression. Pipelining is employed to increase performance and microcoding used to provide flexibility. The design was verified using NASA Prognostics Center of Excellence (PCoE) lithium-ion battery capacity data. Experimental results show that the proposed processor can achieve factors of 7, 2 and 5 improvement in execution time, power and latency over a standard microprocessor solution, while maintaining prediction accuracy. The vector processor is suitable not only for battery capacity prediction, but also for other online time series prediction problems.