Model-free predictive control (MFPC) is an essential robustness control strategy for motor drives. However, due to the impact of stagnation and its negative effects, the accuracy of the adopted data-driven model fails to represent the motion characteristics and operating states of the system. Especially for data-driven models based on data gradient updates such as generalized universal model (GUM), stagnations lead to confusion in the estimation process, posing a threat to system stability.

Exploring anti-stagnation methods is necessary for achieving excellent control performance. By mitigating stagnation, reducing its occurrence probability, and improving the quality of control objective achievement, MFPC constitutes an effective method to meet stringent requirements.

In a study published in IEEE Transactions on Industrial Electronics, Prof. WANG Fengxiang’s group from Fujian Institute of Research on the Structure of Matter of the Chinese Academy of Sciences designed a frequency-converting double second-order generalized integrator (FC-DSOGI) structure to reduce stagnation and its negative effects, and demonstrated the benefits of minimizing stagnation effects and improving current quality by using a GUM-based MFPC as an example for permanent magnet synchronous motors (PMSM) drives.

The signal components generated by parts other than the motor that are not conducive to achieving higher model accuracy goals are called "invalid components." These components have the potential to address stagnation issues. Researchers first analyzed the amplitude-phase characteristics and frequency band distribution of these components, finding them mainly in the high-frequency band with a minor presence in the middle-frequency band, laying the groundwork for their extraction.

Then, researchers designed the FC-DSOGI structure with the motor rotor angular frequency as the resonant center to extract these invalid components. A plateau frequency band for amplitude-frequency characteristics was established based on the invalid components' frequency range. This structure allowed low-frequency and most mid-frequency harmonics to pass through while automatically adjusting the resonant point in real-time according to the motor's current operating state. This ensured the precise acquisition of invalid components without affecting the motor's fundamental wave and harmonic contents.

When the sampled data for the current cycle matched that of the previous cycle, it meant stagnation occurs. The invalid component was inverted, introduced with a random gain, and actively injected back into the sampled data. This created a forced gradient, effectively eliminating stagnation in the data.

Furthermore, researchers conducted a thorough analysis of the system poles and zeros to assess the stability of the GUM-based MFPC. Experimental results confirmed the effectiveness of the MFPC with forced gradients. Compared to the conventional MFPC without anti-stagnation methods, under the same operating conditions, the proposed method exhibited reduced stagnation, improved current quality and higher model accuracy, alongside enhanced robustness. It has compatibility to be applied to other motor drive systems or power electronics topologies to achieve improved control performance. 

This study provides guidance for the future design and development of MFPC strategies that require high-quality predictive accuracy for motor drives.