Genetic algorithms in model structure and controller structure identification

Abstract

Conventional multiloop SISO (single-input, single-output) control systems are still used in industry even though stating interaction between loops often significantly limits their effectiveness. Modern industry requires advanced control solutions based around multivariable system formulations. Two related but specific problems arise in the design of discrete-time MIMO (multi-input, multi-output) control systems. The first is the efficient identification of the structure, order and parameters of the MIMO discrete-time transfer function process description. The second is the difficulty in quickly establishing the selection of an appropriate set of manipulated variables to control a set of specified outputs, often called the ‘pairing problem’. This paper suggests a framework to help solve both problems by the development of automated search procedures based on a genetic algorithm.