Scholarlink

<p>State of Charge (SOC) quantifies stored electricity in batteries, which is crucial for Battery Management Systems (BMS). Despite its significance, SOC cannot be observed directly. Predicting SOC poses challenges owing to the battery’s nonlinear behaviour under complex operating conditions. This paper introduces an Artificial intelligence (AI)-Enabled BMS Hardware Prototype tailored for Electric-powered vehicles (E-Mobility). The SOC estimation method for Lithium-ion batteries employs advanced deep learning techniques, explicitly utilising an LSTM (Long Short-Term Memory) network. Additionally, the paper provides a comprehensive review of research on leveraging AI algorithms and field-programmable gate arrays (FPGAs) to predict and estimate the SOC in lithium-ion batteries (LiBs) used in electric vehicles (EVs). It emphasises the challenges associated with LiBs state estimations and underscores the need for sophisticated AI-based BMS to address them. FPGAs stand out from traditional processors like the Central Processing Unit (CPU) and Graphics Processing Unit (GPU) because they emphasise high-speed acceleration and minimal power consumption. The paper suggests utilising LSTM networks for optimising BMS algorithms and their potential implementation on FPGAs. The paper also explores the implementation of LSTM networks on the Zynq-7020 (XC7Z020) FPGA from Xilinx, demonstrating speed, power and resource utilisation. Additionally, the growing popularity of implementing LSTM on FPGAs is examined due to their low power consumption and fast reconfigurability. The paper offers an in-depth look into implementing LSTM on FPGA devices through High-Level Synthesis (HLS) tools. It thoroughly reviews AI-driven methodologies and hardware accelerators to forecast the SOC in EV lithium-ion Batteries. The proposed technique doesn’t rely on a specific model or knowledge of the battery’s internal parameters. It estimates SOC across different temperatures using a singular set of self-learned network parameters, eliminating the need for temperature-specific models. Given the intricate computational demands of deep learning algorithms, this paper proposes the utilisation of an FPGA-based AI accelerator explicitly designed for E-Mobility BMS. Furthermore, it introduces the development of a corresponding hardware prototype. The Mean Absolute Error (MAE) for Recurrent Neural Network (RNN) stands at 6.87%, while the Gated Recurrent Unit (GRU-RNN) achieves 1.02%. However, the LSTM demonstrates exceptional performance with an average MAE of 0.3274% and a total On-Chip Power consumption of 1.407&#xa0;W.</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

AI-integrated hardware prototype for battery management systems in electric vehicle applications

  • 期刊论文

  • 发表日期：

    2026年3月25日

查看全文
  • Satyashil D. Nagarale,
  • B. P. Patil

摘要

State of Charge (SOC) quantifies stored electricity in batteries, which is crucial for Battery Management Systems (BMS). Despite its significance, SOC cannot be observed directly. Predicting SOC poses challenges owing to the battery’s nonlinear behaviour under complex operating conditions. This paper introduces an Artificial intelligence (AI)-Enabled BMS Hardware Prototype tailored for Electric-powered vehicles (E-Mobility). The SOC estimation method for Lithium-ion batteries employs advanced deep learning techniques, explicitly utilising an LSTM (Long Short-Term Memory) network. Additionally, the paper provides a comprehensive review of research on leveraging AI algorithms and field-programmable gate arrays (FPGAs) to predict and estimate the SOC in lithium-ion batteries (LiBs) used in electric vehicles (EVs). It emphasises the challenges associated with LiBs state estimations and underscores the need for sophisticated AI-based BMS to address them. FPGAs stand out from traditional processors like the Central Processing Unit (CPU) and Graphics Processing Unit (GPU) because they emphasise high-speed acceleration and minimal power consumption. The paper suggests utilising LSTM networks for optimising BMS algorithms and their potential implementation on FPGAs. The paper also explores the implementation of LSTM networks on the Zynq-7020 (XC7Z020) FPGA from Xilinx, demonstrating speed, power and resource utilisation. Additionally, the growing popularity of implementing LSTM on FPGAs is examined due to their low power consumption and fast reconfigurability. The paper offers an in-depth look into implementing LSTM on FPGA devices through High-Level Synthesis (HLS) tools. It thoroughly reviews AI-driven methodologies and hardware accelerators to forecast the SOC in EV lithium-ion Batteries. The proposed technique doesn’t rely on a specific model or knowledge of the battery’s internal parameters. It estimates SOC across different temperatures using a singular set of self-learned network parameters, eliminating the need for temperature-specific models. Given the intricate computational demands of deep learning algorithms, this paper proposes the utilisation of an FPGA-based AI accelerator explicitly designed for E-Mobility BMS. Furthermore, it introduces the development of a corresponding hardware prototype. The Mean Absolute Error (MAE) for Recurrent Neural Network (RNN) stands at 6.87%, while the Gated Recurrent Unit (GRU-RNN) achieves 1.02%. However, the LSTM demonstrates exceptional performance with an average MAE of 0.3274% and a total On-Chip Power consumption of 1.407 W.