Genetic adaptive identification and control

Abstract Genetic algorithms are computer programs that are developed to crudely emulate the evolution of biological populations according to Darwin’s theory of natural selection and the concept of inheritance from genetics put forward by Mendel. Suppose that a controller for a plant is viewed as an individual decision-maker that has been chosen from a population of possible decision-makers to generate a control input to the plant. Suppose that in real-time a population of such decision-makers evolves. The “best” decision-maker from the evolving population is chosen at each step to control the plant, and using the principles of inheritance and survival of the fittest, good decision-makers will be more likely to propagate through the population as it evolves. Generally, as the population evolves and the best decision-maker is chosen at each time step, it adapts to its environment (i.e. the controller adapts to the plant and anything that influences it) and enhanced closed-loop system performance can be obtained. Also, even if there are plant parameter variations or disturbances, the population of decision-makers (controllers) will continually adapt to its environment to try to maintain good performance. This paper discusses a variety of such genetic adaptive control methods, and gives an extensive comparative analysis of their performance relative to conventional adaptive control techniques for an illustrative control application.