A Triple-Integral Noise-Resistant RNN for Time-Dependent Constrained Nonlinear Optimization Applied to Manipulator Control
Yu Han, Guangfeng Cheng, Binbin Qiu
Abstract
Due to the swift advancement of neural networks in recent years, many studies have reported various recurrent neural network (RNN) models aimed at addressing nonlinear optimization problems. However, few of the existing neural networks take time-dependent parameters, inequality constraints, and noise resistance into consideration simultaneously, resulting in the difficulty in solving practical engineering problems. Besides, most methods cannot efficiently suppress time-dependent noise, especially the polynomial noise occurring frequently in practical engineering applications. To tackle this challenge, this article proposes a triple-integral noise-resistant RNN (TINR-RNN) model to efficiently address time-dependent constrained nonlinear optimization (TDCNO) problems limited by multiple equality and inequality constraints under various noise disturbances. The theoretical analyses prove that the residual error generated by the TINR-RNN model can achieve global convergence under multiple noise disturbances, which demonstrates the noise-resistant capability of the TINR-RNN model. Finally, numerical simulation analyses and manipulator control instances substantiate the superior performance of the TINR-RNN model for TDCNO problem solving under various time-dependent noise disturbances, particularly cubic noise.