Intelligent Multi-Objective Nonlinear Model Predictive Control (iMO-NMPC): Towards the ‘on-line’ optimization of highly complex control problems

The benefits of using the Nonlinear Model Predictive Control (NMPC) for the response optimization of highly complex controlled plants are well known. Nevertheless the complexity and associated high computational cost make its implementation and reliability the focus of the discussion. This paper proposes an Intelligent and Multi-Objective NMPC (iMO-NMPC) scheme where several, and often conflicting, control objectives can be competing simultaneously in the control problem. In the iMO-NMPC, the combination of a Neural Network, a Multi-Objective Genetic Algorithm and a Fuzzy Inference System, help us in the nonlinear search for near-optimal control actions, aiming to fulfil all the specified control objectives simultaneously. The proposed scheme adds an expert stage that can adaptively change the degree of importance (weight) of each control objective according to the state of the plant. Therefore, once the nonlinear multi-objective optimization problem is solved at each sampling time and the non-inferior control solutions belonging to the set of Pareto are obtained, the most appropriate one is selected by using the control objectives weights inferred from the expert stage. Some experimental results showing the iMO-NMPC effectiveness and the details about its implementation over control systems with low sampling times are also presented and discussed in this paper.

► We propose an intelligent multi-objective non-linear Model Predictive Control scheme. ► It is a contribution for the on-line optimization of highly complex control problems. ► Three intelligent computational techniques (NN, MOGA, FIS) are combined to solve it. ► At each sampling time the control solutions belonging to the Pareto set are obtained. ► An expert decision stage is used as an adaptive Decision Maker.