On RNN Models for Solving Dynamic System of Linear Equations

Abstract

Neural networks have a wide range of applications in dealing with various online computing problems. This paper mainly retrospects one of the latest recurrent neural network (RNN) models and supplies summarizes on it. Firstly, formulations on the RNN model for dealing with the dynamic underdetermined system of linear equations with double bound constraints on state variables and residual errors are presented. Secondly, simple structures of the RNN model, that is, the neuron-connection architecture of RNN model for handling with the perturbed dynamic underdetermined linear system, as well as the RNN model and the unfolding in time of the computation involved in its forward computation are analyzed. In addition, the whole flowchart on the presented method for establishing the RNN model is also given. Then, experiments on executing the tasks of the UR5 robot when the end-effector tracks a “four-leaf clover” path and a “tricuspid valve” path synthesized by the RNN model are conducted, which show the superiority and accuracy of the presented RNN model.