Global inverse modeling for nonlinear non-affine system control by wavelet network

Abstract

This paper presents a control scheme which learns the inverse mapping of a dynamic system by an orthonormal wavelet network. To compensate the modeling error caused by the model parameterization, feedback is added. The inverse mapping of dynamic system proposed here is defined as a mapping between the output trajectory and input trajectory. Training samples are chosen such that they can cover input trajectory space uniformly both in the amplitude domain and frequency domain. Here the amplitude domain depends on the actuator while the frequency domain depends on sampling period of the control system. For trajectory training, there are a lot of sample data (not sample trajectory) which enhance the complexity of the modeling problem. Hence data compression is used by wavelet threshold which is a method frequently used in signal processing. The performance of the proposed algorithm is illustrated by a computational simulation experiment.