Designing a model predictive controller for graphite baking process

Model predictive controller provides better control quality for graphite products baking process than PID and its variations. The drawback of MPC is its high sensitivity to object modelling errors. If all of object parameters are varied just by 3 % simultaneously then MPC provides satisfactory control no more. In the current paper we design an adaptive MPC for baking process control. In order to identify furnace section “under the fire” on-line, we utilize recursive least squares algorithm with forgetting factor equal to 0,98. We compare conventional and adaptive MPCs in the following cases: accurate model; undesirable deviations of object parameters from model parameters (object is time invariant); desirable deviations; time variant object with step-changing parameters. In each case adaptive MPC provides control that ensures temperature difference inside the hottest (the most “dangerous”) graphite bar to be small enough, and even gives some margin for it (0,8 % to 8 %). However, the cost for this is increased process duration (4,8 % to 142,1 %) and more fuel-consuming (1,2 % to 106,2 %) than in case of non-adaptive MPC, even if the latter provides satisfactory control as well. In general, it is reasonable to use an adaptive MPC instead of a conventional one. Margin of economical reasonability needs further research. Also, it is necessary to compare different identification methods in order to achieve faster transition process.