Experimental validation of modifier adaptation and Gaussian processes for real time optimisation

Plant-model mismatch is a major challenge in the implementation of real time optimisation approaches. Various methods have been proposed to correct for this mismatch based on online data, however there is still little work in the experimental validation and comparison of methods. This work focuses on the experimental implementation and validation of some recently proposed methods to account for plant-model mismatch in a real time optimisation context: output modifier adaption, Gaussian processes, and Gaussian process output modifier adaptation. These methods are implemented on an experimental rig, designed to emulate gas-lifted oil wells drawing from the same reservoir. All the methods are, on average, able to improve upon the baseline performance, with Gaussian process output modifier adaptation showing the best performance. A major challenge was to tune the methods to be robust against the process noise. Copyright (c) 2023 The Authors.