Enhancing PI Tuning in Plant Commissioning through Bayesian Optimization

This paper introduces a new approach aimed at expediting the auto-tuning of PI controllers during the commissioning phase. This approach is based on the exploitation of the process model knowledge and Bayesian optimization capabilities. The process unfolds in a sequence of steps: initially, the quality of the model is improved through the identification of unknown or uncertain parameters of the model. Subsequently, this refined model is used for searching the optimal configuration of PI controller. The outcomes obtained, encompassing the initial estimate, upper and lower bounds, and the Gaussian process mean, are then harnessed to initiate the Bayesian optimization process in the commissioning phase. By initializing adequately the Bayesian optimization, a significant reduction in the number of iterations required to reach the optimizer's optimal solution can be achieved. The approach efficiency is demonstrated through its application to a thermal plant.