ANN-Based Space Vector PWM Modulation for Permanent-Magnet Synchronous Motors

This paper proposes an artificial neural network (ANN)-based space vector PWM (SVPWM) inverter controller for permanent-magnet synchronous motors (PMSM). Traditional SVPWM control methods involve complex computations and exhibit poor robustness to motor parameter variations and load disturbances, ma...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:International Conference on Power Electronics and Drive Systems (Online) s. 1 - 5
Hlavní autoři: Huang, Zhen, Gong, Jiawei, Wang, Chao, Wang, Weiping, Jia, Shaofeng, Huang, Kunjie, Xia, Yonghong
Médium: Konferenční příspěvek
Jazyk:angličtina
Vydáno: IEEE 21.07.2025
Témata:
Aerospace electronics
Artificial neural networks
Backpropagation algorithms
Bayes methods
Computational efficiency
Motors
Robustness
Space vector pulse width modulation
Synchronous motors
Vectors
ISSN:2164-5264
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:This paper proposes an artificial neural network (ANN)-based space vector PWM (SVPWM) inverter controller for permanent-magnet synchronous motors (PMSM). Traditional SVPWM control methods involve complex computations and exhibit poor robustness to motor parameter variations and load disturbances, making them inadequate for high-precision and high-dynamic-response applications. Due to its strong nonlinear mapping capability and adaptability, ANN can optimize SVPWM control strategies, enhancing system real-time performance and robustness. This study employs an ANN trained using the Bayesian regularization backpropagation algorithm and introduces a modular, low-complexity ANN-based SVPWM implementation scheme. Compared to conventional methods, the proposed approach reduces the online computational burden, improves efficiency, and is validated through simulations in the MATLAB/Simulink environment. The results demonstrate that ANN-based SVPWM control maintains high waveform quality across different modulation indices while reducing computational costs by approximately 10 % - 15 %.
ISSN:2164-5264
DOI:10.1109/PEDS63958.2025.11144770