Worst-Case Tolerance Analysis for a Power Electronic System by Modified Genetic Algorithms

It is necessary to consider variations of circuit element values for power electronic converter design to satisfy a given specification for a frequency characteristic or a dynamic transient response because the variations of element values generate those of the designed characteristic. Therefore element values and their variation ranges should be designed to give a satisfactory characteristic within an allowable specification range or a tolerance. This analysis is so-called tolerance or worst-case analysis. The analysis is a sort of nonlinear minimum and maximum (min/max) problem with respect to circuit parameters. However its computation cost is large and practically the problem is converted to a simpler and easier one with various techniques. This paper proposes an approximate method which finds a quasi-min/max solution by applying the genetic algorithm (GA). According to the GA, the higher the fitness value is, the more likely the genetic code remains. The solution is improved by generation and generation efficiently and the solution at the final generation is considered to be optimum. This paper proposes modified GA methods, a sharing-GA and a relay-search-GA, which are suitable for more precise tolerance analysis by utilizing diversity of solutions. Output voltage analysis examples of three types of converters for element parameter variations are investigated