Harmonic-Resistant Recurrent Neural Network for a Time-Varying Matrix Pseudoinverse

The time-varying matrix pseudoinverse holds significant importance across multiple domains. Previous research has indicated that the zeroing neural network model exhibits significant advantages in addressing this issue. However, identified shortcomings include a lack of capability to suppress disturbances effectively. This paper proposes two harmonic-resistant recurrent neural network (HRRNN) models designed to address both known and unknown frequencies. Specifically, these models take into account derivatives of harmonic-type disturbances, effectively suppressing harmonic disturbances. Our focus on the simulated implementation of the HRRNN model presents a novel approach for addressing the challenge of time-varying matrix inversion in low-power scenarios. The effectiveness is validated through numerical simulation experiments.