Scholarlink

<p>An optimal Long Short-Term Memory (LSTM) model combining the Siberia Tiger Optimisation (ST) and Polar Fox Optimization (PF) algorithms is created to enhance the ability of electric vertical take-off and landing (eVTOL) batteries to predict the SoC. To optimise the LSTM network's hyperparameters, the hybrid optimisation platform combines ST global exploration with PF's local exploitation. Models optimized with specific algorithms achieve a mean squared error (MSE) of 0.00034 and a coefficient of correlation (<i>R</i><sup>2</sup>) of 0.9978 Root mean square error (RMSE = 0.01844) when tested using real-time battery datasets; the hybrid LSTM_PF_ST model beats these systems. With slight variation across multiple charge–discharge cycles, a predicted SoC curve fits the real data, indicating an adaptive level of generalization and learning. A more minor prediction error, quicker convergence, and improved performance under dynamic load scenarios are all characteristics of the hybrid model. According to these results, the ST-PF optimised LSTM framework provides a reliable and efficient way to estimate SoC in real time, which supports intelligent energy management, extends flight duration, and ensures operational safety in current electric propulsion 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

Hybrid Deep Learning Framework for Accurate SoC Estimation in eVTOL Systems

  • 期刊论文

  • 发表日期：

    2026年7月29日

查看全文
  • T. Santiago Arockiam,
  • S. Kalimuthu Kumar,
  • Alagar Karthick

摘要

An optimal Long Short-Term Memory (LSTM) model combining the Siberia Tiger Optimisation (ST) and Polar Fox Optimization (PF) algorithms is created to enhance the ability of electric vertical take-off and landing (eVTOL) batteries to predict the SoC. To optimise the LSTM network's hyperparameters, the hybrid optimisation platform combines ST global exploration with PF's local exploitation. Models optimized with specific algorithms achieve a mean squared error (MSE) of 0.00034 and a coefficient of correlation (R2) of 0.9978 Root mean square error (RMSE = 0.01844) when tested using real-time battery datasets; the hybrid LSTM_PF_ST model beats these systems. With slight variation across multiple charge–discharge cycles, a predicted SoC curve fits the real data, indicating an adaptive level of generalization and learning. A more minor prediction error, quicker convergence, and improved performance under dynamic load scenarios are all characteristics of the hybrid model. According to these results, the ST-PF optimised LSTM framework provides a reliable and efficient way to estimate SoC in real time, which supports intelligent energy management, extends flight duration, and ensures operational safety in current electric propulsion systems.