Scholarlink

Abstract <p>This study presents an FPGA-based implementation of an adaptive power management unit (APMU) for electric vehicles (EVs), leveraging a hybrid bidirectional long short-term memory (Bi-LSTM) network and decision tree classifier to optimize power distribution in real time. Developed on the Zynq UltraScale+ MPSoC platform, the proposed system estimates the state of charge (SoC) and classifies driving conditions to dynamically allocate power to onboard components. The FPGA testbed models a mid-range EV by simulating key parameters such as throttle position, vehicle speed, battery voltage/current, and GPS data at 30-second intervals. Experimental results demonstrate significant improvements in power efficiency and computational latency compared to conventional battery management units (BMUs). The proposed system consumes approximately 0.98 W, achieves a latency of 5.6 µs, and operates at 181.6 operations per watt-far surpassing traditional microcontroller or DSP-based BMUs. Range estimation shows up to a 25% increase under highway conditions using the Bi-LSTM + decision tree model, validating the effectiveness of the adaptive strategy for intelligent energy management in EVs.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Enabled Battery Management System on FPGA for Electric Vehicles

  • 期刊论文

  • 发表日期：

    2025年11月5日

查看全文
  • Daisy Merina. R,
  • Saravana Ram. R,
  • Lordwin Cecil Prabhaker Micheal

摘要

Abstract

This study presents an FPGA-based implementation of an adaptive power management unit (APMU) for electric vehicles (EVs), leveraging a hybrid bidirectional long short-term memory (Bi-LSTM) network and decision tree classifier to optimize power distribution in real time. Developed on the Zynq UltraScale+ MPSoC platform, the proposed system estimates the state of charge (SoC) and classifies driving conditions to dynamically allocate power to onboard components. The FPGA testbed models a mid-range EV by simulating key parameters such as throttle position, vehicle speed, battery voltage/current, and GPS data at 30-second intervals. Experimental results demonstrate significant improvements in power efficiency and computational latency compared to conventional battery management units (BMUs). The proposed system consumes approximately 0.98 W, achieves a latency of 5.6 µs, and operates at 181.6 operations per watt-far surpassing traditional microcontroller or DSP-based BMUs. Range estimation shows up to a 25% increase under highway conditions using the Bi-LSTM + decision tree model, validating the effectiveness of the adaptive strategy for intelligent energy management in EVs.