Abstract
A special class of recurrent neural network, termed zeroing neural network (ZNN), has been recently proposed for computing various real-valued time-dependent problems, and it has shown wonderful performance in declining the computing error. When extended to complex-valued domain, ZNN’s characteristics have become complicated. Specifically, the activation function is a tremendous challenge for the complex-valued ZNN. In many previous studies, the linear activation function is adopted, which leads to a slow convergence speed. Adopting an ingenious nonlinear activation function, this work first proposes a new limited-time convergent ZNN (LTCZNN) model for solving time-dependent complex-valued matrix pseudoinverse. Theoretical analyses are conducted, and results show that the LTCZNN model can converge to the optimal solution in the limited time. Moreover, the upper bound of the convergence time is inferred in theory. Finally, the numerical experiment results further confirm those of the theoretical analysis, indicating the superior convergence of the presented LTCZNN model for solving time-dependent complex-valued matrix pseudoinverse.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call