Application Of Immune Algorithm In The Recognition Ofabnormal Spectrum

Immune Clonal Algorithm is a kind of heuristic evolutionary algorithm that is presented recent years, which proposes a new way for complicated problems. It also attracts the attention from researchers who focuses on swarm intelligence. Meanwhile, infrared spectra analysis technique is a kind of fast, lossless and reliable analysis techniques that is developed in1990s, which was called "giant of analysis". Outlier sample recognition infrared spectra analysis guarantees data reliability for building model of infrared spectra analysis. This paper mainly focuses on how to applies Immune Clonal Algorithm to near infrared spec outlier sample recognition by studying the Immune Clonal Algorithm.In this paper, author takes muti threads and concurrence computing into accounts to improve Immune Clonal Algorithm and modifies algorithm to put memory mechanism into it. The modified algorithm not only increases diversity of population but also accelerates the convergence rate greatly. Algorithm is able of doing local search and global search in the process of evolution of antibody population.Firstly, the near infrared spec data of water content, protein content and fat content tested from pork is divided into two parts,1/5of data is used as testing set and other4/5is used as calibration set. Then Concurrent and Immunological memory Clonal selection algorithm, standard Immune Clonal Algorithm, Genetic Algorithm, Mahalanobis distance and Leave-One-Out Cross-Validation Algorithm is used to recognize outlier sample respectively. Remove the outlier sample from calibration set and use the least square method to build model so that we can measure the accuracy of those algorithms. The result is, As for water content, sum of prediction of error square of Concurrent and Immunological memory Clonal selection algorithm is reduced by14.64%,36.64%,55.1%and55.1%by contrast Immune Clonal Algorithm, Genetic Algorithm, Mahalanobis distance and Leave-One-Out Cross-Validation Algorithm; as for fat content, sum of prediction of error square of Concurrent and Immunological memory Clonal selection algorithm is reduced by14.76%,53.49%,89.94%and83.9%by contrast Immune Clonal Algorithm, Genetic Algorithm, Mahalanobis distance and Leave-One-Out Cross-Validation Algorithm; as for protein content, sum of prediction of error square of Concurrent and Immunological memory Clonal selection algorithm is reduced by16.31%,32.91%,52.43%and52.43%by contrast Immune Clonal Algorithm, Genetic Algorithm, Mahalanobis distance and Leave-One-Out Cross-Validation Algorithm. The result shows that Concurrent and Immunological memory Clonal selection algorithm not only recognizes outlier sample data of near infrared spec more accurate, but also speed up the rate of recognizing outlier sample greatly, which ensures the data reliability for the next near infrared spec analysis.