Scholarlink

<p>Accurate State of Charge (SoC) estimation for lithium-ion batteries is vital for managing energy well in electric vehicles and storage systems. This paper introduces a new real-time SoC estimation system. It uses a simple Decision Tree Regression (DTR) model, tested with a Raspberry Pi and an OPAL-RT real-time simulator. Unlike many studies that test offline or use complex models, our method focuses on embedded feasibility, running in real-time, and cost effective. The Raspberry Pi runs the machine learning part, showing that SoC estimation can run on cheap hardware. The OPAL-RT simulator provides real-time voltage and current signals. An MCP3008 analog-to-digital converter picks up these signals, allowing for smooth and clear data exchange between the simulation and the device. This setup, where hardware works with the simulation, lets us test the SoC estimator in conditions that are just like real life. Our tests show the system has a MAE of 0.1427, MAPE of 0.02148, and RMSE of 0.14631. This means it estimates SoC reliably without needing much computational complexity. These results suggest that this method is good for real-time battery management application on embedded devices with limited resources. The novelty of work here is the combination of an efficient machine learning model, cheap embedded hardware, and real-time testing. This bridges the gap between data-driven SoC estimation and practical use in devices. The system also sets up a flexible base for custom changes and future battery management improvements.</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

Real-time battery SoC estimation using machine learning with raspberry Pi and OPAL-RT validation

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Tikam Bhardwaj,
  • Vijay Kale,
  • Makarand Sudhakar Ballal,
  • Sudarshan Khond

摘要

Accurate State of Charge (SoC) estimation for lithium-ion batteries is vital for managing energy well in electric vehicles and storage systems. This paper introduces a new real-time SoC estimation system. It uses a simple Decision Tree Regression (DTR) model, tested with a Raspberry Pi and an OPAL-RT real-time simulator. Unlike many studies that test offline or use complex models, our method focuses on embedded feasibility, running in real-time, and cost effective. The Raspberry Pi runs the machine learning part, showing that SoC estimation can run on cheap hardware. The OPAL-RT simulator provides real-time voltage and current signals. An MCP3008 analog-to-digital converter picks up these signals, allowing for smooth and clear data exchange between the simulation and the device. This setup, where hardware works with the simulation, lets us test the SoC estimator in conditions that are just like real life. Our tests show the system has a MAE of 0.1427, MAPE of 0.02148, and RMSE of 0.14631. This means it estimates SoC reliably without needing much computational complexity. These results suggest that this method is good for real-time battery management application on embedded devices with limited resources. The novelty of work here is the combination of an efficient machine learning model, cheap embedded hardware, and real-time testing. This bridges the gap between data-driven SoC estimation and practical use in devices. The system also sets up a flexible base for custom changes and future battery management improvements.