CityU Institutional Repository >
CityU Electronic Theses and Dissertations >
ETD - Dept. of Electronic Engineering >
EE - Doctor of Philosophy >
Please use this identifier to cite or link to this item:
|Title: ||Clone selection programming with its application|
|Other Titles: ||Ke long xuan ze bian cheng ji qi ying yong|
|Authors: ||Gan, Zhaohui (甘朝暉)|
|Department: ||Department of Electronic Engineering|
|Degree: ||Doctor of Philosophy|
|Issue Date: ||2008|
|Publisher: ||City University of Hong Kong|
|Subjects: ||Immune system -- Computer simulation.|
|Notes: ||CityU Call Number: QR182.2.C65 G36 2008|
xii, 150 leaves : ill. 30 cm.
Thesis (Ph.D.)--City University of Hong Kong, 2008.
Includes bibliographical references (leaves 138-149)
|Abstract: ||Artificial Immune Systems (AIS) represents new bio-inspired algorithms, and it has powerful and robust information processing capabilities for solving complex problems. The application of AIS can be found in classification, clustering, pattern recognition, optimization and fault detection etc. This thesis focuses on the development of new algorithms in the area of AIS and its wide range of applications. We have developed ‘Clone Selection Programming (CSP)’. Using the new CSP model, several real applications are presented to demonstrate that CSP can efficiently be applied in the artificial intelligence and data mining etc. fields.
Based on biological immune system concepts, CSP is an extension of Artificial Immune System (AIS), which is a systematic, domain-independent, and intelligent based method to solve programming problems. In CSP, antibodies represent candidate solutions, which are encoded according to the structure of the antibody. The antibodies are able to keep syntax correct even if they are changed with iterations. The proposed strategies have been thoroughly evaluated by intensive simulations. The results demonstrate the effectiveness and excellent convergent qualities of the CSP based search strategy.
Clone Selection Programming (CSP) is expanded to classification problem, whereby large datasets containing very high dimensional features involving multiple feature-sets can be classified. In a CSP based classifier, antibodies represent candidate solutions of classification rules. The antibodies (classification rules) keep the syntax correct even when they are changed with iterations. Based on the fixed antibody size in CSP, the proposed method overcomes the bloating problems experienced in traditional Genetic Programming (GP). Our proposed master/slave parallel model of multi threading is applied to learn a group of rules. A post processing method is used to improve classification performance. The classification performance using the CSP model is compared to conventional methods.
Combined with novel entropy based clustering validity index, CSP was applied to the automatic clustering of unlabeled datasets. Being different from most existing clustering techniques, the proposed algorithm does not require any a priori knowledge of data to partition the dataset. It can find the optimal number of clusters and group data into an optimal partitioning dynamically. The effectiveness of the CSP based clustering algorithm is demonstrated by clustering several synthetic and real-life datasets with the number of dimensions ranging from two to ten，and the number of clusters ranging from two to nine.
Clonal Selection Programming (CSP) can also be used in industrial applications. A fault detection system was developed for performing induction machine fault detection and analysis. Four feature vectors are extracted from power spectra of machine vibration signals. The extracted features are inputs of a CSP-based classifier for fault identification and classification. The proposed CSP-based machine fault diagnostic system has been intensively tested with unbalanced electrical faults and mechanical faults operating at different rotating speeds. The proposed system is not only able to detect electrical and mechanical faults correctly, but the rules generated are also very simple and compact and easy for people to understand.
A Clonal Selection Programming (CSP) based nonlinear system modeling method is developed and presented. The proposed method uses a multi gene encoding scheme to represent regressors of the nonlinear system. Each gene consists only of the functional operators multiplication and time delay, and link between genes being an addition function. The proposed method does not require any a priori information about the system, i.e., dimension, maximum lag degree, nonlinear degree, etc. The proposed CSP approach uses the Least-Squares (LS) method to determine the parameters of the nonlinear system models. In this study, the CSP method was evaluated by modeling noisy polynomial systems. Simulation results show that the newly developed CSP method is an effective alternative for performing nonlinear system identification. The presented results show that the CSP approach can consistently deliver accurate and compact nonlinear models compared with other conventional and evolutionary methods.|
|Online Catalog Link: ||http://lib.cityu.edu.hk/record=b2340673|
|Appears in Collections:||EE - Doctor of Philosophy |
Items in CityU IR are protected by copyright, with all rights reserved, unless otherwise indicated.