Genetic algorithms in multiobjective control designs
Student thesis: Master's Thesis
Related Research Unit(s)
|Award date||30 Jul 1999|
This thesis presents the use of Genetic Algorithms (GA) in the multiobjective control design problems. The theme is to devise the means and methods for such multiobjective problems based upon the intrinsic characteristics of GA. Control engineering problems often exist in the class of multiple objectives. In the conventional techniques, these problems are usually converted into a single objective problem or handled in a similar manner. The major deficiency of those approaches is that only a single best solution is located. It is not sufficient in most of the cases since a set of optimal solutions, known as the Pareto optimal set, intrinsically exists in this special class of problems. GA is a stochastic searching algorithm inspired by the mechanics of natural selection and genetic. The goodness of the chromosome (or the solution candidate) in GA depends on its fitness value instead of the corresponding objective values. All the GA operations are performed accounting to this fitness value. With such intrinsic characteristic together with the concept of Pareto set, it is possible to formulate the multiobjective GA (MOGA) for solving the complex multiobjective control problems in a way that other methods have failed to meet. The fitness value of chromosome is defined as its rank in the Pareto set. With such definition, MOGA can locate multiple solutions with highest rank and eventually identify a set Pareto optimals even in a single run. To demonstrate the effectiveness of the MOGA, it is applied to two control design problems. Firstly, it is adopted to design a fuzzy logic controller(FLC) for the oil flowing system of a solar plant. It takes into account of the competing objectives and optimizes simultaneously the membership functions and fuzzy control rules of the FLC. An optimal solution is then selected from the obtained Pareto set by the decision maker to obtain satisfactory control results. The MOGA is also used to tackle with a benchmark problem which is a time-varying plant with stringent performance criteria. The MOGA is to design the optimal pre-compensator and post-compensator of the H∞, robust controller using the Loop Shaping Design Procedure. The system requirements in different stressed environments are entirely met by the use of the multiobjective approach.
- Automatic control, Genetic algorithms