Adaptive predictive control using GOBF-ARX models: An experimental case study

Industrial applications of model predictive control rely mostly on linear empirical models obtained by employing time series analysis approaches. These models can quickly become obsolete and require maintenance when the operating conditions become significantly different from the design conditions. The need to generate good predictions in the face of changing operating conditions and/or plant characteristics can be fulfilled through updating the linear model parameters online. This work is aimed at the development of adaptive MPC (AMPC) scheme based on ARX models, which are parameterized using generalized orthonormal basis filters (GOBF). The proposed model structure, in addition to capturing the dynamics with respect to the manipulated inputs, facilitates modeling of stationary as well as non-stationary components of the unmeasured disturbances. The feasibility of using the proposed AMPC scheme is established by conducting experimental studies on a benchmark Heater-Mixer setup.