Noise-tolerant Z-type neural dynamics for online solving time-varying inverse square root problems: A control-based approach

Abstract

Assuming that the solving process is free of measurement noises, the zeroing neural dynamics (Z-type) based on different zeroing dynamics for online solving time-varying inverse square root problems (TVISRPs) are revisited from the perspective of control technique and unified into a control framework. However, noises are ubiquitous and unavoidable in the application of the real-time system. Therefore, the modified Z-type models (MZTMs) with different measurement noises are needed for online solving TVISRPs. In this study, the MZTMs are first developed, analyzed and verified for online solution of the TVISRPs with different measurement noises. Furthermore, theoretical analyses infer that the MZTMs globally and exponentially converge to the theoretical solution. Compared with the traditional Z-type neural dynamics model (ZTNDM) and gradient neural dynamics model (GNDM), two illustrative examples are described and investigated to substantiate the efficiency and superiority of the developed MZTMs. Finally, a systematic method is proposed via exploring control framework to construct MZTMs for online solving TVISRPs with great robustness and accuracy.