Designing Optimal Operational-Point Trajectories Using an Intelligent Sub-strategy Agent-Based Approach

Abstract

This paper presents a method intended for designing optimal and safe control for nonlinear dynamical processes. The sought control signal results from elementary control strategies induced by different agents implementing their (partial) task of minimizing a common control cost measure (index). The issue of designing optimal control is therefore treated as a decision process, where the decisions are made in particular regions of the state space of the dynamical process under consideration. The regions thus constitute local decision spaces being searched by a group of agents in a multistage searching procedure. At each stage, every agent can increment its cost index only by a limited value. This guarantees that at the end of each stage all the agents represent control strategies which are cost equivalent (approximately). The algorithm starts off by generating an initial population of agents (each for one of the previously defined elementary control strategies). Each of these agents realizes a different kind of possible elementary control strategies, which determine predefined agent behaviors. When an agent reaches one of the decision regions, it generates a new/local population of the seeking/hunting agents (they are, again, of different kinds of the elementary control strategies). After getting explored, such a decision region turns to a forbidden zone for all agents but those belonging to the newly created population. In such a way the successive populations of the agents allow to complete the path to a prearranged destination point in a competitive way. The first agent which reaches the destination area in the state space determines an optimal solution in the sense of the above assumptions.