Aiming at the shortage of premature convergence in the conventional genetic programming
a new feature generation method based on immune programming is put forward
where the new features are constructed by polynomial expressions of the original features
and the antibody affinity is defined by the ratio of inter-class scatter to the intra-class scatter. Combining cloning
hypermutation and replacement operator
the clonal selection optimal algorithm is adopted to search the best feature with excellent classification performance. The experiment of the six time domain features for engine sound signal shows that due to the maintained diversity of antibodies based on the clonal selection principle
the best compound feature constructed by immune programming is endowed with the better classification ability than the feature optimized by genetic programming.
关键词
Keywords
references
Vafaie H, de Jong K. Feature space transformation using genetic algorithms[J]. IEEE Intelligent System,1998,13(2):57-65.
Chen Peng, Nasu M, Toyota T. Self-reorganization of symptom parameters in frequency domain for failure diagnosis by genetic algorithm[J]. Journal of Intelligent and Fuzzy Systems,1998,6(1):27-37.
Li Liangmin,Qu Liangsheng. Genetic programming for dimensionless parameters construction[J]. Journal of Xi'an Jiaotong University,2005,36(7):736-739.
Durga P, Nikhil R, Jyotirmoy D. Genetic programming for simultaneous feature selection and classifier design[J]. IEEE Trans on System,Man,and Cybernetics, B: Cybernetics,2006,36(1):106-117.
Sun Ruixiang,Fugee T. Combing bootstrap and genetic programming for feature discovery in diesel engine diagnosis[J]. International Journal of Industrial Engineering,2004,11(3):273-281.
Peter M, Adriel L, Marek R, et al. Immune programming[J]. Information Sciences,2006,176(8):972-1002.
Koza J R, Streeter M J, Keane M. Routine high-return human-competitive machine learning[C]∥Proceedings of the International Conference on Machine Learning and Applications. Boston:Kluwer Academic Publishers,2003: 6-12.