Multiple layer perceptron for Direct Inverse Control of a nonlinear system

The idea of using neural networks in control of dynamic system is comparatively new. The well-known and well-established control methodologies such as classical control (PID) and modern control techniques were developed for control of linear systems. However many practical systems are nonlinear. Nonlinear methods for control do exist but vary from case to case. Neural networks offer a simple and extremely flexible solution to problem of control of nonlinear systems. This paper proposes direct inverse control scheme of a simple nonlinear dynamic system. The idea is to train a neural network as an inverse of plant so as to cancel out the plant dynamics and to make plant follow the reference input.