On the Use of Variable Mutation in an Evolutionary Algorithm

When an evolutionary algorithm is used to optimize a function or process it performs a broad search of solution space. To find the probable optimum point the evolutionary algorithm must be complimented by a local search technique. Creep mutation is one such possible technique. Specifying suitable mutation control parameter values that suit both the evolutionary algorithm and the local search requirements is not easy, especially if the solution space is complex and little detail is known about it. This paper describes the use of cyclically varying mutation control parameter values to allow creep mutation to be successful over a wide range of parameter values.