Component selection optimization using genetic algorithms.

摘要

Selecting components for multi-energy systems that have transient performance specifications requires a wide range of knowledge, including system design, modeling, dynamic analysis, computer programming, and optimization. The purpose of this research was to reduce the time required to design and optimize a system that has transient performance design criteria by providing a design method to automate the selection of an optimal set of components for a given layout.; The component selection method developed in this thesis consists of a genetic algorithm, developed especially for this type of design problem, a generalized constraint handling procedure, and recommendations as to the method's uses. A genetic algorithm is a flexible, evolutionary, combinatorial optimization technique that lends itself well to problems with discrete solution spaces. The genetic algorithm developed in this work uses a population retention scheme that retains good designs from one generation to the next, uses uniform crossover as the mating scheme, and uses a Russian roulette population reduction that systematically eliminates poor designs from the pool of potential solutions. A generalized penalty function approach was developed that does not require foreknowledge about the design problem's solution. This penalty function is dependent on time; at the start of the algorithm the penalty for violating a design constraint is lenient, but increases with time. This approach is very general and can be used for linear or non-linear constraints. Running multiple, independent copies of this algorithm either in parallel or serially is highly recommended. By running the algorithm multiple times, the probability of obtaining an optimal answer is greatly increased.; Two different industrial design problems were used successfully to test the component selection method. This method finds a good design quickly without data conditioning or using rules such as an expert system would require.