Exploring the Inherent Fault-Tolerance of Model-Free Predictive Control in Three-Level T-Type Converters

Fault-tolerant control (FTC) improves the reliability of three-level T-type converters under open-circuit faults in power devices. However, existing FTC methods rely on comprehensive and precise fault diagnosis, which significantly increases the complexity of diagnostic algorithms and limits their adaptability. To address this issue, this work proposes a highly adaptive FTC method based on model-free predictive control using an ultra-local model. The proposed approach dynamically estimates model disturbances caused by faults by employing a neural network observer. This eliminates the need for additional control variations due to different fault scenarios, and thus enables a unified control strategy. Furthermore, this method only requires the prior identification of vertical device faults without the need to diagnose horizontal devices or pinpoint specific faulty components, thereby reducing the complexity of fault diagnosis. Comparative experiments validate the proposed method’s adaptive fault-tolerant performance and robustness against filter parameter variations under various fault conditions.