Rule extraction of fault diagnose based on a Modified artificial immune algorithm

摘要

When employed in fault diagnosis, rough set can realize attribution reduction. But it can not discrete attribution and reduct attribution simultaneously, therefore we can not say that it can automatically extract rules. To solve the problem, a new rule extraction method based on developed artificial immune algorithm is firstly proposed in the paper. At first, a new method of encoding is produced which can make the process of discretion and reduction unify. Secondly, a new definition of concentration of antibodies not only compare individuals in structure and space, but also in fitness value. Thirdly, the algorithm provide dissimilation operator and similar-taxis operator, which replace choice, expansion and mutation in traditional artificial immune algorithm. All these developments not only maintain diversity of the antibody population, but also converge faster. Finally, we apply the algorithm to fault diagnose of heat recoup system in steam turbine. Tests proved that the algorithm is feasible, and the diagnose rules acquired by the algorithm have higher accuracy rate.