Feature Clustering and Fault-Tolerant Control of Multimodal Missing Data in a Fully Actuated System

The Fully Actuated System refers to a system in which the control inputs are capable of driving all state variables, thereby achieving global control of the system. However, traditional control methods often face certain limitations when dealing with complex systems, especially when the system exhibits strong coupling, nonlinearity, or uncertainty. To address the issue of missing data in fully actuated systems and better reveal the nonlinear relationships between variables in different controllers, this paper proposes a novel fully actuated system multimodal missing data feature clustering and fault-tolerant control method–Anchor-point-based Consistency Multimodal Space Contrastive Prediction Model(APC). In this model, the anchor points are first selected through a self-suppressing random walk method, and the data is re-represented based on the anchor points. A reconstruction loss function is used to constrain the extracted latent features, ensuring their effective representation of variable information. At the same time, the Kullback-Leibler (KL) divergence method is employed to maximize the mutual information between the two modalities, ultimately predicting and completing the missing data based on the similarity of the multimodal data. Experimental results show that the proposed model demonstrates good effectiveness in addressing the multimodal missing data problem in fully actuated systems.