An online PI auto-tuning method is proposed for superheat control for a type of supermarket refrigeration systems. The proposed procedure consists of three serial steps: Step-One uses one of the two proposed empirical methods, namely
multi-step method and relay method, for modeling initialization. The objective is to guide the initial (superheat) operation into a neighborhood of the predefined set-point and keep the operation inside afterwards. Step-Two identifies a FOPDT model for the concerned local dynamic either through a Modified Area Method (MAM) or a Least-Square Prediction (LSP) method. A PI controller is auto-tuned based on the obtained FOPDT model and the SIMC method in Step-Three. The proposed method is implemented on a real-sized physical system and the experimental results showed a promising potential to apply the proposed method into commercial development.