.png){kind=icon}
Return
Highly Adaptable Data-Physics-Driven Predictive Control Using Likelihood Weights on PMSM Drives
.png){kind=icon}
PRE
AI
DOI：10.1109/jestpe.2026.3734433.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Model-free predictive control (MFPC) replaces physical models with data-driven models to realize essential robustness. However, especially in the case of parameter matching, the data-driven model usually lacks high adaptability, which is different from physical models built based on the plant’s operation mechanisms. To improve adaptability, a data-physics-driven predictive control (DPDPC) using likelihood weights and a model silence mechanism is proposed in this paper, and adopted as the current controller for the permanent magnet synchronous motor (PMSM) drives. The plant model integrates a physical model with multiple data-driven models to generate flexible control signals using a set of adaptive weights. Derived from the probability distribution of sampled data, a likelihood function is designed to dynamically select weights, ensuring the most adaptable model dominates, and models with continuously smaller weights will become silent to generate more additional resources and prevent overrun errors. Theoretical analysis confirms the stability of the system under different weights. The experimental results validate the effectiveness of the proposed method, indicating that compared to MFPC using ultra-localized autoregressive with exogenous input (ARX), the proposed method has improved current quality and model adaptability. Furthermore, the proposed method exhibits enhanced robustness, although it is not totally model-free.
Keywords:
Data-physics-driven predictive control (DPDPC)
likelihood weights
high adaptability
permanent magnet synchronous motor (PMSM) drives
Journal
I
IF:
4.9
Papers:
369
Citations:
0
Organization
No organization information available
Cited Papers
No cited papers available