Implementation of Extremum Seeking Control in an Experimental Lab-Rig

Abstract

Extremum seeking control (ESC) is an adaptive control method that can be used for production optimization. In the standard ESC approach, the process optimum is found by slowly adapting the inputs based on plant gradient approximations. Since ESC “learns” the system characteristics online using only plant measurements, it requires an appropriate time-scale separation between the system response, the excitation signal used for gradient approximation, and the dynamics to be optimized. Krishnamoorthy et al. (2019) proposed a dynamic ESC that identifies a local linear dynamic model using transient measurements and applies this model to compute the plant gradients. Consequently, the time-scale separation between the system response and the excitation signal is not required and the scheme converges to the optimum faster than the standard ESC. Albeit attractive, some practical challenges have been identified, such as gradient computation and constraint handling. The main contribution of this paper is to investigate how to overcome these challenges in a more realistic setting, in which dynamic ESC is implemented on a lab-scale plant that emulates a subsea oil well network. We compare the performance of two gradient computation approaches and present a constraint handling strategy for the system of interest, where multiple units compete for limited resources. The results show that dynamic ESC is able to drive the system to its optimum without constraint violations.