A Cooperative Learning Approach to Qualitative Mixed Performance Controllers

摘要

In this paper, based on the idea of cooperative learning, a novel method to design a family of mixed performance optimal controllers is presented. Each controller is assumed as an expert agent that is professional in one field —in accordance to its design objectives. A Q-table-like tabular form is introduced to represent the knowledge and expertise of each controller agent. Then, a cost function, which is a combination of initial controllers' cost functions, is defined. The mixed performance controller is designed to minimize this new cost function. A closed form for the mixed performance controller is found. Proposed method is illustrated in a design example, where applicability of the method is justified.