Empirical investigation and analysis of the computational potentials of bio-inspired nonlinear model predictive controllers: success and challenges

Abstract

In this investigation, a comprehensive study is carried out to excavate the potentials of bio-inspired computing (BIC) for the development of model predictive controllers (MPCs) for different classes of nonlinear problems. The two mentioned fields are now playing pivotal roles in industry, and there is a large consensus on the fact that BIC and MPCs are among the most applicable techniques in the coming decades. One of the most important decisions for developing MPCs is the selection of the optimisation technique. Here, the authors would like to demonstrate the applicability of BICs to be used as an optimisation method at the heart of MPCs to calculate the controlling commands. The resulting controllers are applied to some challenging problems to clearly demonstrate the applicability of BICs for developing high-performance MPCs.

Recent Meta-Heuristics Improved by Self-Adaptation Applied to Nonlinear Model-Based Predictive Control

“…The MPC performs the control by calculating the optimal actions based on the predicted future behavior of a given plant through an online optimization process [1]. Gradient descent algorithms have been used for the purpose of solving these optimization problems [2]- [4] but it has been shown to be limited in computational capacity for nonlinear and non-convex problems [5]. Regarding this, Bioinspired Meta-heuristics (BMs) proved to be an alternative for dealing with these complex problems [6], for example in advanced control [7], [8].…”

“…BMs' ability to explore and exploit in linear, nonlinear, and multi-objective problems makes it useful in various classes of MPC, including linear MPC, nonlinear MPCs (NMPCs), tracking-based MPCs, hybrid MPCs, among others. This fact presents another advantage of BMs over Traditional Algorithms (TAs) since the latter has the primary ability to solve convex optimization problems [5]. One of the challenges in designing and implementing NMPC in a real-world context is to seek algorithms capable of obtaining the optimal control actions for a given system since different models require different solutions.…”