Parameters Tuning of an Immune Algorithm for Time-Series Pattern Recognition to Improve Ability to Escape from a Local Optimum and Achieve a Better Solution

Abstract

In this paper, time-series pattern recognition is applied for a damage evaluation sequence represented by a frequency response change due to a damage coefficient occurrence. The objective of this paper is parameters tuning of pattern recognition system which discovers dependencies between data entries in time-series. The influence of input parameters over the ability of the immune algorithm to identify similarity of historical sequences describing frequency response change of an object (antibodies) as a result of a damage coefficient occurrence to a sequence describing frequency response with a noise due to external disturbances (pathogen) is researched. The investigated parameters are: affinity threshold At, suppression threshold St, number of mutated genes Nm, number of iterations Ino and number of clones K. Antibodies (sequences) belonging to a given neighborhood represent a given type of damage. Antibodies from the neighborhood undergo clonal selection and affinity maturation process to achieve better affinity to the pathogen. With the best matched antibody the type of damage in the problem of time-series pattern recognition is achieved.