Model Predictive Controller Tuning in Electrical Distribution Grids by Multiobjective Optimization

Abstract

In this thesis a multiobjective tuning of a linear state signal shaping model predictive control (LSSSMPC) controller is implemented in an electrical distribution grid to provide power quality services. The tuning is done using the evolutionary algorithm Multiobjective Differential Evolution with Spherical Pruning X (spMODEx). Two research questions are addressed. In the first stage the reproducibility of the spMODEx algorithm, when approximating the Pareto Front, as the solution to the multiobjective problem statement, is addressed. The assessment is realized by means of singular value decomposition, and additionally by a novel approach the activated volume method, first reported herein. In the second stage a decision making strategy is developed to find the best controller parameters in a reasonable amount of time. Controller and plant are operated in a mismatch. The controller operates on a conventional linear state space model, while the plant simulation is done by a four mode switched system to adequately model the behavior of non–linear loads. The thesis concludes with the outcomes on the reproducibility of spMODEx algorithm, the activated volume method, and recommendations for a successful multiobjective optimization and decision making process to ensure tuning parameters that allow the LSSSMPC controller to provide optimal power quality services.