Towards stable neural network controllers

This paper presents a neural network approach for the stable control of unknown nonlinear systems, which are defined on a non-compact set. The basic idea is to transfer the non-compact definition domain of a nonlinear function to a bounded open set. This transfer is achieved via a one-to-one mapping. It is shown that B-splines neural networks can then be trained and used for controller design. A stable closed-loop system is therefore obtained. A simulation example is included, which demonstrates the use of the new method and obtains the desired results.