基于ZYNQ深度学习模型部署的锂电池健康预测

Lithium-ion Battery Health Prognostic Based on a ZYNQ Deep Learning Model

为实现低成本、快速、高精度的电动汽车锂电池健康预测,提出一套基于深度学习模型和ZYNQ硬件平台的锂电池健康预测系统。首先,使用锂电池历史退化数据训练多层卷积神经网络模型,得到模型各层的权重;其次,采用动态精度数据量化策略对各层的模型参数进行量化,有效减少内存占用和带宽需求,进而提升硬件加速器的性能;然后,利用高层次综合工具实现了模型参数在ZYNQ硬件平台的嵌入和部署;最后,整套系统在Xilinx XC7020平台上进行验证。实验结果表明,该系统在实现高精度预测的同时,能够有效降低功耗、降低时延,满足车载条件下锂电池健康管理系统的嵌入式应用需求。

In order to realize low-cost, fast and high-precision health prognostics of lithium-ion batteries for electric vehicles, a system for lithium-ion battery health prognostic based on a deep learning model and a ZYNQ hardware platform is proposed in this paper. Firstly, a multi-layer convolutional neural network model is pretrained by using the historical degradation data of lithium-ion batteries, and the model weights in each layer are preserved. Secondly, the model weights are quantified by using a dynamic precision data quantization strategy,which can effectively reduce the memory consumption and bandwidth requirements, and improve the performance of the hardware accelerator. Thirdly, the quantified model weights are embedded and deployed into a ZYNQ hardware platform by using the high-level synthesis tool. Finally, the whole system is validated on the Xilinx XC7020 platform. The experimental results reveal that the proposed system can effectively reduce power consumption and time delay while achieving a high precision prognostic, and meet the requirements of embedded applications for lithium-ion battery health management system in electric vehicles.