Scholarlink

<p>Permanent magnet synchronous motors (PMSMs) are widely used in electric vehicles (EVs) due to their high efficiency, fast dynamic response, and stable operation. Model predictive control (MPC) has shown a strong potential for PMSM speed regulation by explicitly handling system constraints and delivering optimal control actions. However, its high computational complexity limits its practical deployment on resource constrained embedded platforms. To address this, in this paper, a deep neural network–based MPC (DNN-MPC) strategy is proposed, where a neural network learns the MPC policy from offline-generated data. A significant challenge in imitation learning is the absence of formal constraint guarantees, which can compromise the safe operation of the motor. To address this, a novel training framework incorporating logarithmic barrier functions into the loss function has proposed to enable strict adherence to the system constraints. A Luenberger observer is employed for state estimation, enhancing robustness against measurement noise. The proposed approach is validated through hardware-in-the-Loop (HIL) co-simulations, demonstrating that the proposed DNN-MPC achieves performance comparable to classical MPC while significantly reducing computational time and memory demands. The results confirm that the framework ensures safe, efficient, and real-time speed control of the PMSM motor, making it suitable for resource-constrained automotive embedded implementations.</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

Constraint-guaranteed Embedded DNN-MPC for PMSM Drives

  • 期刊论文

  • 发表日期：

    2026年2月26日

查看全文
  • Nirlipta Ranjan Mohanty,
  • Pramod Ubare,
  • Vaishali Patne,
  • Dayaram Sonawane

摘要

Permanent magnet synchronous motors (PMSMs) are widely used in electric vehicles (EVs) due to their high efficiency, fast dynamic response, and stable operation. Model predictive control (MPC) has shown a strong potential for PMSM speed regulation by explicitly handling system constraints and delivering optimal control actions. However, its high computational complexity limits its practical deployment on resource constrained embedded platforms. To address this, in this paper, a deep neural network–based MPC (DNN-MPC) strategy is proposed, where a neural network learns the MPC policy from offline-generated data. A significant challenge in imitation learning is the absence of formal constraint guarantees, which can compromise the safe operation of the motor. To address this, a novel training framework incorporating logarithmic barrier functions into the loss function has proposed to enable strict adherence to the system constraints. A Luenberger observer is employed for state estimation, enhancing robustness against measurement noise. The proposed approach is validated through hardware-in-the-Loop (HIL) co-simulations, demonstrating that the proposed DNN-MPC achieves performance comparable to classical MPC while significantly reducing computational time and memory demands. The results confirm that the framework ensures safe, efficient, and real-time speed control of the PMSM motor, making it suitable for resource-constrained automotive embedded implementations.