Study On Artificial Immune Algorithm And Its Application To Intelligent Fault Diagnosis Of Marine Diesel Engine

In this dissertation, a thorough study on artificial immune algorithm (AIA) is given. By combining AIA and computational intelligence fault diagnosis approach, an intelligent diagnosis method for marine diesel engine is deeply researched. The main achievements are included as follows:(1) First, some basic concepts, functions and principles of the biological immune system are discussed. Then the basic theory, structure and process of the AIA are analyzed. Based on the analysis on opt-aiNet algorithm's theory and performance, an improved immune algorithm for multi-modal functions optimization is proposed through a new network suppress method named as valley searching.(2) In order to achieve cluster analysis with unknown number of clusters, a fast immune dynamic clustering algorithm is put forword, which is inspired by the clone selection principle of the vertebrate immune system and combines the cluster validity analysis. It not only adaptively determines the amount and the center's positions of clustering, but also avoids the local optima. The convergence speed of this algorithm is improved obviously through introducing a new search operation and selecting appropriate initial clustering center.(3) Base on thorough study on principle and method of RBF neural network, a new study algorithm is proposed, which can automatically determine the number and positions of hidden layer RBF centers by the fast immune dynamic clustering algorithm, and the weights of output layer are decided by the recursive least squares algorithm. A basic principle of fault diagnosis neural network-based is given, and fault diagnosis for marine diesel engine is achieved.(4) The core content of rough sets theory is introduced and a novel attribute reduction algorithm of rough set based on the immune optimization is proposed. The key of this algorithm is to integrate discernible ability and the elements in the condition attribute set into one unified affinity maturation objection, and to maintain the diversity of antibody population with renewal of antibody and similar antibodies suppression. So the different attribute reduction sets can be found, which can maintain the ability of classification.