Scholarlink

<p>The precise estimation of battery states like the state of charge (SOC) and state of power (SOP), is essential to maintain the reliability, safety, and efficiency of battery management system (BMS) in electric vehicles (EVs). This study presents a extensive comparison between a feed forward neural network (FNN) and a deep feed forward neural network (DFNN) for the prediction of SOC and SOP in lithium ion 18650 29E energy and 18650 25R power cells. SOC was determined using open circuit voltage (OCV) linear interpolation method and SOP from the profile of instantaneous voltage and current. This study presents a data centric framework to estimate the SOP of lithium ion cells using experimentally obtained pulse power characteristics. Pulse discharge tests (1 s to 240 s) across the SOC range demonstrate nonlinear SOP degradation at low SOC. Polynomial and exponential models precisely represent this behaviour, with the second order polynomial delivering optimal accuracy and computational efficiency for real time BMS implementation. In addition, range estimation, scenario based power requirement, minimum SOC requirements and power consumption analysis was conducted. The results indicate that the FNN model demonstrates substantially higher reliability and prediction accuracy compared to the DFNN, reliably delivering more precise SOC estimations for both power and energy cells. In comparison, the DFNN exhibits more significant deviations in prediction performance thus indicating that merely increasing the model depth does not always result into higher accuracy. In addition, architectures such as FNN can attain superior SOC predictive performance. Furthermore, the developed characterisation experimental procedure differentiates between functional and non functional cells such that only healthy functional cells are identified to build battery packs which further enhances the safety feature for EVs. The proposed framework provides a flexible and effectively implementable solution for future generation intelligent battery management systems.</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 based neural network framework for state of charge and state of power mapping estimation of lithium-ion cells for electric vehicles

  • 期刊论文

  • 发表日期：

    2026年4月15日

查看全文
  • Nitesh K A,
  • Ravichandra,
  • P K Ashwini Kumar

摘要

The precise estimation of battery states like the state of charge (SOC) and state of power (SOP), is essential to maintain the reliability, safety, and efficiency of battery management system (BMS) in electric vehicles (EVs). This study presents a extensive comparison between a feed forward neural network (FNN) and a deep feed forward neural network (DFNN) for the prediction of SOC and SOP in lithium ion 18650 29E energy and 18650 25R power cells. SOC was determined using open circuit voltage (OCV) linear interpolation method and SOP from the profile of instantaneous voltage and current. This study presents a data centric framework to estimate the SOP of lithium ion cells using experimentally obtained pulse power characteristics. Pulse discharge tests (1 s to 240 s) across the SOC range demonstrate nonlinear SOP degradation at low SOC. Polynomial and exponential models precisely represent this behaviour, with the second order polynomial delivering optimal accuracy and computational efficiency for real time BMS implementation. In addition, range estimation, scenario based power requirement, minimum SOC requirements and power consumption analysis was conducted. The results indicate that the FNN model demonstrates substantially higher reliability and prediction accuracy compared to the DFNN, reliably delivering more precise SOC estimations for both power and energy cells. In comparison, the DFNN exhibits more significant deviations in prediction performance thus indicating that merely increasing the model depth does not always result into higher accuracy. In addition, architectures such as FNN can attain superior SOC predictive performance. Furthermore, the developed characterisation experimental procedure differentiates between functional and non functional cells such that only healthy functional cells are identified to build battery packs which further enhances the safety feature for EVs. The proposed framework provides a flexible and effectively implementable solution for future generation intelligent battery management systems.