Scholarlink

<p>This study presents a real-time deployable pre-trained adaptive controller for double-sided linear induction motors. A deep surrogate model is employed for online gain scheduling, while a population-based optimizer refines controller parameters within a 200-microsecond sampling interval. The surrogate is trained offline using 15,000 operating scenarios incorporating electrical, magnetic, and mechanical parameter variations. Controller performance is evaluated across twelve operating scenarios using both simulation and OPAL-RT hardware-in-the-loop experiments. The observed average overshoot is 3.2%, the maximum overshoot remains below 10%, and the steady-state error remains below 0.2% under the tested conditions. Statistical analysis using Shapiro–Wilk, Levene, ANOVA or Friedman tests with Holm-adjusted post hoc comparisons indicates statistically significant performance differences relative to benchmark methods (<i>p</i> &lt; 0.01). Frequency response analysis based on locally linearized models reports gain and phase margins of approximately 6&#xa0;dB and 45 degrees, respectively. The proposed approach demonstrates reduced computational effort compared to a model predictive control baseline. Overall, the results indicate that surrogate-assisted metaheuristic control is experimentally feasible for real-time DSLIM operation within the evaluated operating conditions.</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 adaptive control of double-sided linear induction motors using deep surrogate-assisted metaheuristic optimization

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Varaha Narasimha Raja Chilukoti,
  • S. Sandeep Kumar Reddy Vaka,
  • Murali Vantaku,
  • Baseem Khan,
  • Rajkumar Sivanraju,
  • Khaled S. Alatawi,
  • Fahad M. Almasoudi,
  • Ahmed Ali

摘要

This study presents a real-time deployable pre-trained adaptive controller for double-sided linear induction motors. A deep surrogate model is employed for online gain scheduling, while a population-based optimizer refines controller parameters within a 200-microsecond sampling interval. The surrogate is trained offline using 15,000 operating scenarios incorporating electrical, magnetic, and mechanical parameter variations. Controller performance is evaluated across twelve operating scenarios using both simulation and OPAL-RT hardware-in-the-loop experiments. The observed average overshoot is 3.2%, the maximum overshoot remains below 10%, and the steady-state error remains below 0.2% under the tested conditions. Statistical analysis using Shapiro–Wilk, Levene, ANOVA or Friedman tests with Holm-adjusted post hoc comparisons indicates statistically significant performance differences relative to benchmark methods (p < 0.01). Frequency response analysis based on locally linearized models reports gain and phase margins of approximately 6 dB and 45 degrees, respectively. The proposed approach demonstrates reduced computational effort compared to a model predictive control baseline. Overall, the results indicate that surrogate-assisted metaheuristic control is experimentally feasible for real-time DSLIM operation within the evaluated operating conditions.