Real-Time Tuning Using System Identification for Climate Control System

Sammanfattning

This thesis investigates the possibility of deriving a controller tuning procedure based on black-box modelling with closed-loop system identification. As many users operate their systems 24/7, they are neither able nor willing to shut down for retuning procedures as operating conditions change. Thereby, the ability to adjust control parameters in real-time becomes increasingly important. To generate informative data, input excitation signals like steps, chirps and telegraphs were applied. The direct approach to closed-loop system identification was used to estimate black-box models of the ARX, ARMAX and BJ structures and methods like observing pole/zero diagrams and coefficient uncertainties were used to validate models. First order with delay models were approximated from the best performing black-box models and were used to compute new control parameters. Performance metrics like the Harris index and squared control error were calculated before and after retuning to evaluate the effectiveness of the retuning. The telegraph input excitation signal generated informative data both in noisy simulations as well as on a physical test unit. From evaluation of the black-box structures, the BJ structure consistently performed the best, accurately capturing the system dynamics. Utilising the noise models from this structure to calculate the Harris index indicated better controller performance after applying the new control parameters. With these results in mind, a tuning framework could be built. These results imply that the tuning procedure suggested by this thesis is well founded. It can deliver new control parameters that help the system track the operating setpoint. Future work should emphasise automating the tuning procedure as well as calculating the Harris index over time to indicate when a retuning is necessary.