Parameters selection of LSSVM based on adaptive genetic algorithm for ship rolling prediction

Abstract

The Least Squares Support Vector Machines (LSSVM) is a promising artificial intelligence technique in which the regression algorithm has already been used solve the nonlinear function approach successfully. The nuclear function parameter and penalty parameter is a pivotal factor which decides performance of LSSVM. Unfortunately most users selected parameters for an LSSVM by rule of thumb, so they frequently fail to generate the optimal approaching effect for the function. This has restricted effective use of LSSVM to a great degree. To solve these problems, a new approach based on an adaptive genetic algorithm (AGA) was proposed, which automatically adjusts the parameters for LSSVM, this method selects crossover probability and mutation probability according to the fitness values of the object function, therefore reduces the convergence time and improves the precision of genetic algorithm (GA), insuring the accuracy of parameter selection. This method was applied to ship rolling prediction, and simulation results showed it can effectively improve prediction accuracy.