Adaptive Smith Predictive Control of Non-Linear Systems Using Neurofuzzy Hammerstein Models

This paper proposes an Adaptive Smith Predictor Controller (ASPC) based on Neuro-Fuzzy Hammerstein Models (NFHM) with on-line non-linear model parameters identification. The NFHM approach uses a zeroorder Takagi-Sugeno fuzzy model to approximate the non-linear static function that is tuned off-line using gradient decent algorithm and to identify the linear dynamic function it is used the Recursive Least Square estimation with Covariance Matrix Reset (RLSCMR). This algorithm has the capability of follow fast and slow dynamic parameter changes. The proposed ASPC has special capabilities to control non-linear systems that have gain, time delay and dynamic changes through time. The implementation of the ASPC is made in two steps: first, off-line estimation of the non-linear static parameters that will be used to “get linear” the non-linearity of the system and second, on-line identification of the linear dynamic parameters updating direct and inverse models used in the ASPC. As an illustrative example, a gas water heater system is controlled with the ASPC. Finally, the control results are compared with the results obtained with the Smith Predictive Controller based in a Semi-Physical Model (SPMSPC). 1 This work was supported in part by the Portuguese Government through the PREDEP program.