Model-free　predictive　control　(MFPC)　operates　entirely　independently　of　physical　models　and　parameters　by　leveraging　a　data-driven　model.　However,　the　processes　of　modeling　and　updating　these　models　demand　high-quality　sampled　data,　with　stagnation　and　its　negative　effect　posing　significant　obstacles　to　the　advancement　of　this　technology.　To　overcome　this　challenge,　this　paper　proposes　an　anti-stagnation-based　MFPC　specifically　for　permanent　magnet　synchronous　motor　(PMSM)　drives.　This　approach　features　a　notch　structure　designed　to　extract　specific　frequency　band　harmonics　generated　by　the　control　strategy　and　then　inversely　inject　them　into　the　sampled　data　to　create　data　gradients.　This　method　aims　to　minimize　the　risk　of　stagnation　and　alleviate　its　adverse　effects,　while　also　incorporating　control　compensation.　At　the　theoretical　level,　a　comprehensive　analysis　of　the　method＇s　stability　and　robustness　is　conducted.　Experimental　results　show　that,　compared　to　the　strategy　without　an　anti-stagnation,　the　proposed　method　offers　superior　current　quality　and　prediction　accuracy,　providing　a　novel　and　effective　solution　for　high-performance　control　of　PMSM　drives　in　complex　environments.　©　2025　IEEE.