Synthesis of feedback linearization and variable structure control with neural net compensation

The synthesis of feedback linearization and variable-structure control (VSC) for uncertain systems with an artificial neural network (ANN) as feedforward compensation is considered. Feedback linearization provides a systematic approach such that an appropriate coordinate system can be found in which the system can be linearized. The VSC is designed to produce a desired overall behavior in spite of existing system uncertainties. When the ANN is trained using several patterns from a class of reference inputs, the ANN remembers the equivalent control effort of VSC when the control system is in sliding mode. Appealing to the interpolation property of the ANN, by combining the ANN as a generalized feedforward compensation, the unnecessary high gain of VSC near the equilibrium can be greatly reduced