Implementations of a Hammerstein fuzzy-neural model for predictive control of a lyophilization plant

This paper describes two methodologies for implementation of Hammerstein model by using different input-output representations into model predictive control schemes. The model nonlinearity is easily approximated using a simple Takagi-Sugeno inference, while the linear parts are flexibly introduced. As optimization procedures for predictive control are used a standard gradient optimization method and an implementation of Hildreth Quadratic Programming. A comparison between the proposed control strategies is made by simulation experiments for control of nonlinear lyophilization plant.