混合智能计算方法及其应用
|
摘要
本论文主要以机器学习的两个基本问题——模式识别和函数逼近为背景，对基于神经网络、模糊逻辑、进化算法、免疫算法、量子算法、基于核的学习机等软计算方法的若干混合智能计算方法进行了研究，并将这些混合方法应用于图像处理、目标识别以及系统辨识等具体问题中。主要工作概括如下：
1．首先以求解一些NP问题，如TSP问题为例，讨论了运行在量子计算机上的量子搜索算法和运行在经典计算机上的进化搜索算法之间的本质区别。接着以背包问题为例，定性分析了通过这两种搜索算法之间的结合而形成的量子驱动的进化算法(简称量子进化算法)的特性。通过分析可以看到，量子进化算法结合了量子计算的一些概念和理论，诸如量子比特和量子叠加态，因此能够表示出解的线性叠加态，并且由于量子比特染色体的概率表示，它比传统的进化算法具有更好的种群多样性和全局寻优能力。
2．提出一种自适应免疫进化算法，并用于图像的分割。该算法能够从其最佳个体的基因中自适应地提取有效信息制作成免疫疫苗。同时在接种疫苗的操作中，引入一个自适应变化的参数用来表示接种疫苗个体的百分比，这个参数随着代数而递增，最后增大到1，这意味着所有个体都接受接种疫苗，这样在进化的后期，算法就以大量的局部爬山搜索为特征。该算法用于图像分割问题时，产生了令人满意的分割结果，并对噪声有较好的抑制作用。
3．将免疫算子的概念结合到量子进化算法中，提出一种免疫量子进化算法。免疫量子进化算法在保留原有量子进化算法优良特性的前提下，力图有选择、有目的地利用待求问题中的一些特征信息或先验知识，抑制或避免求解过程中的一些重复或无效的工作，以提高算法的整体性能。对背包问题的仿真结果表明，与标准进化算法、免疫进化算法、量子进化算法相比，免疫量子进化算法不仅是有效的，也是可行的，并能进一步改善算法的性能，提高其收敛速度。同时我们将该算法用于图像的边缘检测问题也得到了较好的检测结果。
4．基于小波变换和进化神经网络提出一种混合的雷达一维距离像的目标识别方法，即首先利用小波变换作为特征提取器，从小波变换后的系数中提取雷达目标特征，然后基于一混合进化算法优化设计了一个前向网络并以此作为分类器对提取的模式特
1叮刁邑电心卜月呼书眺人学们卜创七学亡比今仑李忆
征进行识别。该算法能同时优化网络的拓扑结构和连接权值，并且由于其全局搜索能
力，可以避免结构的局部最小。实验结果表明，用该算法设计出的雷达目标神经网络
分类器结构紧凑，具有较好的泛化能力。
5.研究了基于核的学习算法进行目标识别的方法，该方法把无监督学习(基于核
的主分量分析算法的特征提取)和有监督学习(基于近似支撑矢量机的分类)结合起
来。核的主分量分析算法可以有效地提取出目标中的非线性特征;而近似支撑矢量机
作为一种新型的支撑矢量机，不需要求解二次规划问题，只需对一个简单的线性方程
系统求解，可以快速地对目标进行分类。该方法应用于雷达目标一维像的识别时，其
正确识别率与标准支撑矢量机相当，但在计算速度上却有很大的提高，并对噪声具有较
好的抑制作用。
6.提出了两种非线性系统的混合辨识方法。第一种方法方法首先在结构辨识中，
通过聚类算法自动地从已知的输人输出数据中生成一个初始的模糊模型;接着在参数
辨识中，设计了一个模糊神经网络，通过有监督学习逐步调整网络的权值，也即模糊
模型的参数以使模型具有更高的精度。在第二种方法中提出一种新型混合神经网络模
型一自适应模糊神经网络，该网络结构简洁，具有通用逼近的能力，采用有师学习和
无师学习相结合的混合算法进行训练，收敛速度快。并且受学习样本空间分布的影响
较小，因而有较强的学习能力和表达能力。
The integration of different learning and adaptation techniques, to overcome individual limitations and achieve synergetic effects through hybridization or fusion of these techniques, has in recent years contributed to a large number of new intelligent system designs. This dissertation is focused on the solving the series of hybrid intelligent computing methods based on soft computing technologies consisting of neural networks, fuzzy logic systems, evolutionary algorithms, immune algorithms, quantum algorithms and kernel-based learning machines, and applying these hybrid methods to some practical problems, such as image processing, target recognition and nonlinear system identification.
The main contents of the dissertation are summarized as follows:
1. We firstly compare how a quantum search algorithm running hi a quantum computer differs from an evolutionary search algorithm running on a classical computers for solving NP problems by using the instance of TSP. Then, we take the knapsack problem as an example to make qualitative analysis of the characteristics of quantum-inspired evolutionary algorithm (QEA) through combination of these two techniques. It is shown that QEA can represent linear superposition of states since it is based on the concept and principles of quantum computing such as quantum bit and linear superposition of states. In addition, QEA has a better characteristic of diversity and global search than classical approaches due to its probabilistic representation.
2. An adaptive immune evolutionary algorithm (AIEA) is proposed, and is applied to the image segmentation problem. AIEA can adaptively extract useful information from genes of the current optimal individual and make vaccines during evolution. At the same time, AIEA introduces an adaptive tuned parameter to denote the fraction of individuals in the current population to be subjected to the vaccination operation. This parameter is incremented by a small amount after each generation. Eventually it is equal to 1, which means all individuals have vaccination. So the very late stages of AIEA are characterized by a large number of local hill-climbing moves. Experimental results show that AIEA performs well in terms of quality of the final segmented image and robustness to noise.
3. Immune concepts and methods are led into quantum-inspired evolutionary algorithm (QEA), and a novel algorithm, the immune quantum evolutionary algorithm (IQEA) is presented. On condition of preserving QEA's advantages, it utilizes some characteristics and knowledge in the pending problems for restraining the" repeat and ineffective operations during evolution, so as to improve the algorithm efficiency. The experimental results on the knapsack problem show that the performance of IQEA is superior to the conventional evolutionary algorithm, the immune evolutionary algorithm and QEA. IQEA is also used to the problem of edge detection, and obtains satisfactory results.
4. Based on the combination of the wavelet transform and an evolutionary neural network, we introduce a hybrid approach for radar target recognition by the range profiles. We firstly employ the wavelet transform to extract and select features from the high feature space by taking into account the non-stationary characteristic of the radar echoes. Then, we design a feed-forward neural network as classifier by using a hybrid evolutionary algorithm based on evolutionary programming. This hybrid algorithm can evolve very compact network structure, and the network classifier thus has good generalization ability.
5. A fast method for radar identification by range profiles is proposed based on the kernel algorithms. The whole recognition process consists of two stages. The first is concerned with feature extraction where the kernel principal component analysis is used to select the nonlinear features of range profiles. The second is concerned with pattern classification where the proximal support vector machine is constructed as classifier. Experiment results indicate that the propos
1．首先以求解一些NP问题，如TSP问题为例，讨论了运行在量子计算机上的量子搜索算法和运行在经典计算机上的进化搜索算法之间的本质区别。接着以背包问题为例，定性分析了通过这两种搜索算法之间的结合而形成的量子驱动的进化算法(简称量子进化算法)的特性。通过分析可以看到，量子进化算法结合了量子计算的一些概念和理论，诸如量子比特和量子叠加态，因此能够表示出解的线性叠加态，并且由于量子比特染色体的概率表示，它比传统的进化算法具有更好的种群多样性和全局寻优能力。
2．提出一种自适应免疫进化算法，并用于图像的分割。该算法能够从其最佳个体的基因中自适应地提取有效信息制作成免疫疫苗。同时在接种疫苗的操作中，引入一个自适应变化的参数用来表示接种疫苗个体的百分比，这个参数随着代数而递增，最后增大到1，这意味着所有个体都接受接种疫苗，这样在进化的后期，算法就以大量的局部爬山搜索为特征。该算法用于图像分割问题时，产生了令人满意的分割结果，并对噪声有较好的抑制作用。
3．将免疫算子的概念结合到量子进化算法中，提出一种免疫量子进化算法。免疫量子进化算法在保留原有量子进化算法优良特性的前提下，力图有选择、有目的地利用待求问题中的一些特征信息或先验知识，抑制或避免求解过程中的一些重复或无效的工作，以提高算法的整体性能。对背包问题的仿真结果表明，与标准进化算法、免疫进化算法、量子进化算法相比，免疫量子进化算法不仅是有效的，也是可行的，并能进一步改善算法的性能，提高其收敛速度。同时我们将该算法用于图像的边缘检测问题也得到了较好的检测结果。
4．基于小波变换和进化神经网络提出一种混合的雷达一维距离像的目标识别方法，即首先利用小波变换作为特征提取器，从小波变换后的系数中提取雷达目标特征，然后基于一混合进化算法优化设计了一个前向网络并以此作为分类器对提取的模式特
1叮刁邑电心卜月呼书眺人学们卜创七学亡比今仑李忆
征进行识别。该算法能同时优化网络的拓扑结构和连接权值，并且由于其全局搜索能
力，可以避免结构的局部最小。实验结果表明，用该算法设计出的雷达目标神经网络
分类器结构紧凑，具有较好的泛化能力。
5.研究了基于核的学习算法进行目标识别的方法，该方法把无监督学习(基于核
的主分量分析算法的特征提取)和有监督学习(基于近似支撑矢量机的分类)结合起
来。核的主分量分析算法可以有效地提取出目标中的非线性特征;而近似支撑矢量机
作为一种新型的支撑矢量机，不需要求解二次规划问题，只需对一个简单的线性方程
系统求解，可以快速地对目标进行分类。该方法应用于雷达目标一维像的识别时，其
正确识别率与标准支撑矢量机相当，但在计算速度上却有很大的提高，并对噪声具有较
好的抑制作用。
6.提出了两种非线性系统的混合辨识方法。第一种方法方法首先在结构辨识中，
通过聚类算法自动地从已知的输人输出数据中生成一个初始的模糊模型;接着在参数
辨识中，设计了一个模糊神经网络，通过有监督学习逐步调整网络的权值，也即模糊
模型的参数以使模型具有更高的精度。在第二种方法中提出一种新型混合神经网络模
型一自适应模糊神经网络，该网络结构简洁，具有通用逼近的能力，采用有师学习和
无师学习相结合的混合算法进行训练，收敛速度快。并且受学习样本空间分布的影响
较小，因而有较强的学习能力和表达能力。
The integration of different learning and adaptation techniques, to overcome individual limitations and achieve synergetic effects through hybridization or fusion of these techniques, has in recent years contributed to a large number of new intelligent system designs. This dissertation is focused on the solving the series of hybrid intelligent computing methods based on soft computing technologies consisting of neural networks, fuzzy logic systems, evolutionary algorithms, immune algorithms, quantum algorithms and kernel-based learning machines, and applying these hybrid methods to some practical problems, such as image processing, target recognition and nonlinear system identification.
The main contents of the dissertation are summarized as follows:
1. We firstly compare how a quantum search algorithm running hi a quantum computer differs from an evolutionary search algorithm running on a classical computers for solving NP problems by using the instance of TSP. Then, we take the knapsack problem as an example to make qualitative analysis of the characteristics of quantum-inspired evolutionary algorithm (QEA) through combination of these two techniques. It is shown that QEA can represent linear superposition of states since it is based on the concept and principles of quantum computing such as quantum bit and linear superposition of states. In addition, QEA has a better characteristic of diversity and global search than classical approaches due to its probabilistic representation.
2. An adaptive immune evolutionary algorithm (AIEA) is proposed, and is applied to the image segmentation problem. AIEA can adaptively extract useful information from genes of the current optimal individual and make vaccines during evolution. At the same time, AIEA introduces an adaptive tuned parameter to denote the fraction of individuals in the current population to be subjected to the vaccination operation. This parameter is incremented by a small amount after each generation. Eventually it is equal to 1, which means all individuals have vaccination. So the very late stages of AIEA are characterized by a large number of local hill-climbing moves. Experimental results show that AIEA performs well in terms of quality of the final segmented image and robustness to noise.
3. Immune concepts and methods are led into quantum-inspired evolutionary algorithm (QEA), and a novel algorithm, the immune quantum evolutionary algorithm (IQEA) is presented. On condition of preserving QEA's advantages, it utilizes some characteristics and knowledge in the pending problems for restraining the" repeat and ineffective operations during evolution, so as to improve the algorithm efficiency. The experimental results on the knapsack problem show that the performance of IQEA is superior to the conventional evolutionary algorithm, the immune evolutionary algorithm and QEA. IQEA is also used to the problem of edge detection, and obtains satisfactory results.
4. Based on the combination of the wavelet transform and an evolutionary neural network, we introduce a hybrid approach for radar target recognition by the range profiles. We firstly employ the wavelet transform to extract and select features from the high feature space by taking into account the non-stationary characteristic of the radar echoes. Then, we design a feed-forward neural network as classifier by using a hybrid evolutionary algorithm based on evolutionary programming. This hybrid algorithm can evolve very compact network structure, and the network classifier thus has good generalization ability.
5. A fast method for radar identification by range profiles is proposed based on the kernel algorithms. The whole recognition process consists of two stages. The first is concerned with feature extraction where the kernel principal component analysis is used to select the nonlinear features of range profiles. The second is concerned with pattern classification where the proximal support vector machine is constructed as classifier. Experiment results indicate that the propos
引文
[1] Mjolsness E. and Decoste D. Machine learning for science state of the art and future prospects. Science, 2001,293: 2051-2055．
[2] Mitchell T. Machine Learning. New York: McGraw-Hill, 1997．
[3] Zadeh L. A. Roles of soft computing and fuzzy logic in the conception, design and deployment of information/intelligent systems. Computational Intelligence: Soft Computing and Fuzzy-Neuro Integration with Applications, 1998: 1-9．
[4] Jang J. S. R. ANFIS: Adaptive network based fuzzy inference system. IEEE Trans, on System, Man and Cybernetics, 1993,23 (3) : 665-684．
[5] Jang J. S. R. Neuro-fuzzy modeling: architecture, analysis and application. Ph D Dissertation, University of California, Berkey, 1992．
[6] Jang J. S. R. and Sun C. T. Neurofuzzy modeling and control. Proceedings of IEEE, 1995, 83:378-406．
[7] Halgamuge S. K. and Glesner M. Fast perception learning by fuzzy controlled dynamic adaption of neural parameters. Fuzzy Systems in Computer Science, 1994: 129-139．
[8] Narazaki H. and Ralescu S. A synthesis method for multi-layered neural network using fuzzy sets. Proceedings of the International Workshop on Fuzzy Logic in Artificial Intellience, 1995: 315-334．
[9] Cherkassky V. Fuzzy inference system. Computational Intelligence: Soft Computing and Fuzzy-Neuro Integration with Applications, Springer, 1998:230-259．
[10] Medsker L. R. Hybrid Intelligent Systems. Kluwer Academic Publishers, 1995．
[11] Kandel A. and Langholz G. Hybrid Architecture for Intelligent System, CRC Press, 1992．
[12] Abraham A. and Nath B. Hybrid intelligent systems design: a review of a decade research, http://www-mugc.cc.monash.edu.au/-abrahanip/publication.html.
[13] Nauck D. and Kruse R. A neuro fuzzy controller learning by fuzzy error propagation.
Proceedings of Conference of North American Fuzzy Information Processing Society, Mexico, 1992:388-397．
[14] Herrera F., Lozano M. and Veregay L. Tuning fuzzy logic control by genetic algorithm. International Journal of Approximate Reasoning, 1995,12(3/4) : 1414-1427．
[15] Mohammadian M. and Staonier R. J. Generating fuzzy rules by genetic algorithms. Proceedings of 3rd IEEE International Workshop on Robot and Human Communication, Nagoya, 1994:362-367．
[16] Hoffmann F. and Pfister G. Automatic design of hierarchical fuzzy controllers using genetic algorithm. 2nd European Congress on Intelligent Techniques and Soft Computing, 1994．
[17] Furuhashi T. Development of if-then rules with the use of DNA coding. Fuzzy Evolutionary Computation, Kluwer Academic Publishers, 1997．
[18] Hoffmann F. Soft computing techniques for the design of mobile robot behaviors. Journal of Information Sciences, 2000,122(2-4) : 241-258．
[19] Nauck D., Klawonn F. and Kruse R. Foundations of Neuro-Fuzzy Systems. Wiley, 1997．
[20] Nauck D. and Kruse R. New learning strategies for NEFCLASS. Proceedings of Seventh International Fuzzy systems Association World Congress, Vol. IV, 1997: 129-139．
[21] Tano S., Oyama T. and Arnould T. Deep combination of fuzzy inference and neural network in fuzzy inference. Fuzzy Sets and Systems, 1996,28(2) : 151-160．
[22] Oyama T., Tano S. et al. FINEST: fuzzy inference Environment software with tuning. Proceedings of IEEE International Conference on Fuzzy Systems, 1995: 3-4．
[23] Juang C. F. and Lin C. T. An online self constructing neural fuzzy inference network and its applications. IEEE Trans, on Fuzzy Systems, 1998, 6(1) : 12-32． .
[24] Sulzberger S. M., Tscbicholg-Gurman N. N. and Vestli S. J. FUN: optimization of fuzzy rule based systems using neural networks. Proceedings of IEEE Conference on Neural Networks, San Francisco, 1993: 312-316．
[25] Kasabov N. Evolving fuzzy neural networks-algorithms, applications and biological motivation. Methodologies for the Conception. Design and Application of Soft Computing, World Scientific, 1998:271-274．
[26] Kasabov N. and Qun S. Dynamic evolving fuzzy neural networks with 'm-out-of-n' active nodes for online adaptive systems. Technical Report Tr99/04, Department of information science, University of Otago, 1999．
[27] Dasgupta D. Artificial Immune Systems and Their Applications. Spring-Verlag, 1999．
[28]Hunt J. E., Cooke D. E. Learning using an artificial immune system. Journal of Network and Computer Applications, 1996, 19:189-212.
[29]Tazawa I., Koakutsu S. and Hirata H. An evolutionary optimization based on the immune system and its application to the vlsi floor-plan problem. Trans of the Institute of Electrical Engineers of Japan (Series C and in Japanese), 1997, 117(7): 821-828.
[30]Tazawa I., Koakutsu S. and Hirata H. An optimization method based on the immune system. Proceedings of World Congress on Neural Networks, International Neural Network Society 1996 Annual Meeting, San Diego, CA, USA, 1996: 1045-1049.
[31]王磊，潘进，焦李成．基于免疫策略的进化算法．自然科学进展，2000，10(5)：451-455．
[32]王磊．免疫进化计算理论及应用．西安电子科技大学博士学位论文，2001．
[33]周家驹，何险峰译．演化程序—遗传算法和数据编码的结合．北京：科学出版社，2000．
[34]Kitana H. Empirical studies on the speed of convergence of the neural network training by genetic algorithm. Proceedings of AAAI-90, 1990.
[35]Whitley D., Starkweather T. and Bogart C. Genetic algorithms and neural networks: optimizing connections and connectivity. Parallel Computation, 1990, 14(3): 347-361.
[36]Whitley D. The GENITOR algorithm and selective pressure: why rank-based allocation of reproductive trails is best. Proceedings of 3rd International Conferernce of Genetic Algorithms and Their Applications, San Mateo, CA: Morgan Kaufmnn, 1989:116-121.
[37]Caudell T. P. and Dolan C. P. Parametric connectivity: training of constrained networks using genetic algorithms. Proceedings of 3rd International Conference of Genetic Algorithms and Their Applications, San Mateo, CA: Morgan Kaufmnn, 1989:370-374.
[38]Srinivas M. and Patnalk L. M. Learning neural network weights using genetic algorithms--improving performance by search-space reduction. Proceedings of IJCNN'91, Singapore, Vol. 3, 1991: 2331-2336.
[39]Garis H. GenNets: genetically programmed neural networks—using the genetic algorithm to train neural nets whose inputs and/or outputs vary in time. Proceedings of IJCNN'91, Singapore, Vol.2, 1991: 1391-1396.
[40]Janson D. J. and Frenzel J. F. Application of genetic algorithm to the training of higher order networks. Journal of System Engering, 1992, 2:272-276.
[41]Janson D. J. and Frenzel J. F. Training product unit neural networks with genetic algorithms. IEEE Expertise, 1993, 8:26-33.
[42] Montana D. and Davis L. Training feedforward neural networks using genetic algorithms Proceedings of 11th International Joint Conference on Artificial Intelligence. San Mateo, CA: Morgan Kaufmann, 1989:762-767．
[43] Bartlett P., and Downs T. Training a neural network with a genetic algorithm. Technical Report, Department of Electric Engineering, University of Queensland, Australia, 1990．
[44] Menczer F. and Parisi D. Evidence of hyperplanes in the genetic learning of neural networks. Biological Cybernetics, 1992, 66: 283-298．
[45] Skinner A. J. and Broughton J. Q. Neural networks in computational materials science: training algorithms. Modeling and Simulation in Materials Science Engineering, 1995, 3(3) : 371-391．
[46] Fogel D. B., Fogel L. J. and Parto V. W. Evolving neural networks. Biological Cybernetics, 1990,63(6) : 487-493．
[47] Greenwood G. W. Training partially recurrent neural networks using evolutionary strategies. IEEE Trans, on Speech Audio Processing, 1997, 5(1) : 192-194．
[48] Saravanan N. and Fogel D. B. Evolving neural control systems. IEEE Expertise, 1996, 10:23-27．
[49] Fogel D. B., Wasson E. C. and Porto V. W. A step toward computer-assisted mammography using evolutionary programming and neural networks. Cancer Letter, 1997,118(1) : 93-105．
[50] Stork D. G. et al. Preadaptation in neural circuits. Proceedings of International Joint Conference on Neural Networks, vol. 1, Washington DC, 1990:202-205．
[51] White D. and Ligomenides P. GANNet: a genetic algorithm for optimizing topology and weights in neural network design. IWAANN'93, Berlin: Springer-Verlag, 1993:322-327．
[52] Liu Y. and Yao X. Evolutionary design of artificial neural networks with different nodes. Proceedings of ICEC'96: Nagoya, Japan, 1996, 670-675．
[53] Oliker S., Furst M. and Maimon O. A distributed genetic algorithm for neural network design and training. Complex Systems, 1992, 6(5) : 459-477．
[54] Marin F. J. and Sandoval F. Genetic synthesis of discrete-time recurrent neural network. IWANN'93, vol. 686, Berlin: Springer-Verlag, 1993:179-184．
[55] Alba E et al. Fully automatic.ANN design: a genetic approach. IWANN'93, vol. 686, Berlin: Springer-Verlag, 1993:399-404．
[56] Fang J. and Xi Y. Neural network design based on evolutionary programming. Artificial Intelligent Engineering, 1997, 11(2) : 155-161．
[57] Maniezzo V. Genetic evolution of the topology and weight distribution of neural networks. IEEE Trans, on Neural Networks, 1994, 5(1) : 39-53．
[58] Yao X., Liu Y. A new evolutionary system for evolving artificial neural networks. IEEE Trans, on Neural Networks. 1997, 8 (3) : 694-713．
[59] Yao X. and Liu Y. Towards designing artificial neural networks by evolution. App. Math. Computation, 1998,91(1) : 54-65．
[60] Greenwood G. W. Training partially recurrent neural networks using evolutionary strategies. IEEE Trans, on Speech Audio Processing. 1997, 5(1) : 192-194．
[61] Kitano H. Designing neural networks using genetic algorithms with graph generation system. Complex Systems, 1990,4(4) : 461-476．
[62] Merrill J. W. J. and Port R. F. Fractally configured neural networks. Neural Networks, 1991,4(1) : 53-60．
[63] Narayanan A. and Moore M. Quantum-inspired genetic algorithm. Proceedings of IEEE International Conference on Evolutionary Computation, 1999: 61-66．
[64] Han K. H. and Kim J. H. Genetic quantum algorithm and its application to combinatorial optimization problem. Proceedings of the 2000 IEEE Congress on Evolutionary Computation, 2000: 1354-1360．
[65] Han K. H., Park K. H., Lee C. H. and Kim J. H. Parallel Quantum-inspired genetic algorithm for combinatorial optimization problem. Proceedings of the 2001 IEEE Congress on Evolutionary Computation, 2001: 1442-1429．
[66] Cortes C. and Vapnik V. N. Support vector networks. Machine Learning, 1995,20:273-297．
[67] Williams C. K. I. Computation with infinite neural networks. Neural Computation, 1998, 10(5) : 1203-1216．
[68] Bennett K. P. and Blue J. A. A support vector machine approach to decision trees. Proceedings of DCNN'98, Anchorage, Alaska, 1997: 2396-2401．
[69] Bennett K. Combining support vector and mathematical programming methods for induction. Advances in Kernel Methods-SV Learning, Cambridge, MA, MIT Press, 1999:307-326．
[2] Mitchell T. Machine Learning. New York: McGraw-Hill, 1997．
[3] Zadeh L. A. Roles of soft computing and fuzzy logic in the conception, design and deployment of information/intelligent systems. Computational Intelligence: Soft Computing and Fuzzy-Neuro Integration with Applications, 1998: 1-9．
[4] Jang J. S. R. ANFIS: Adaptive network based fuzzy inference system. IEEE Trans, on System, Man and Cybernetics, 1993,23 (3) : 665-684．
[5] Jang J. S. R. Neuro-fuzzy modeling: architecture, analysis and application. Ph D Dissertation, University of California, Berkey, 1992．
[6] Jang J. S. R. and Sun C. T. Neurofuzzy modeling and control. Proceedings of IEEE, 1995, 83:378-406．
[7] Halgamuge S. K. and Glesner M. Fast perception learning by fuzzy controlled dynamic adaption of neural parameters. Fuzzy Systems in Computer Science, 1994: 129-139．
[8] Narazaki H. and Ralescu S. A synthesis method for multi-layered neural network using fuzzy sets. Proceedings of the International Workshop on Fuzzy Logic in Artificial Intellience, 1995: 315-334．
[9] Cherkassky V. Fuzzy inference system. Computational Intelligence: Soft Computing and Fuzzy-Neuro Integration with Applications, Springer, 1998:230-259．
[10] Medsker L. R. Hybrid Intelligent Systems. Kluwer Academic Publishers, 1995．
[11] Kandel A. and Langholz G. Hybrid Architecture for Intelligent System, CRC Press, 1992．
[12] Abraham A. and Nath B. Hybrid intelligent systems design: a review of a decade research, http://www-mugc.cc.monash.edu.au/-abrahanip/publication.html.
[13] Nauck D. and Kruse R. A neuro fuzzy controller learning by fuzzy error propagation.
Proceedings of Conference of North American Fuzzy Information Processing Society, Mexico, 1992:388-397．
[14] Herrera F., Lozano M. and Veregay L. Tuning fuzzy logic control by genetic algorithm. International Journal of Approximate Reasoning, 1995,12(3/4) : 1414-1427．
[15] Mohammadian M. and Staonier R. J. Generating fuzzy rules by genetic algorithms. Proceedings of 3rd IEEE International Workshop on Robot and Human Communication, Nagoya, 1994:362-367．
[16] Hoffmann F. and Pfister G. Automatic design of hierarchical fuzzy controllers using genetic algorithm. 2nd European Congress on Intelligent Techniques and Soft Computing, 1994．
[17] Furuhashi T. Development of if-then rules with the use of DNA coding. Fuzzy Evolutionary Computation, Kluwer Academic Publishers, 1997．
[18] Hoffmann F. Soft computing techniques for the design of mobile robot behaviors. Journal of Information Sciences, 2000,122(2-4) : 241-258．
[19] Nauck D., Klawonn F. and Kruse R. Foundations of Neuro-Fuzzy Systems. Wiley, 1997．
[20] Nauck D. and Kruse R. New learning strategies for NEFCLASS. Proceedings of Seventh International Fuzzy systems Association World Congress, Vol. IV, 1997: 129-139．
[21] Tano S., Oyama T. and Arnould T. Deep combination of fuzzy inference and neural network in fuzzy inference. Fuzzy Sets and Systems, 1996,28(2) : 151-160．
[22] Oyama T., Tano S. et al. FINEST: fuzzy inference Environment software with tuning. Proceedings of IEEE International Conference on Fuzzy Systems, 1995: 3-4．
[23] Juang C. F. and Lin C. T. An online self constructing neural fuzzy inference network and its applications. IEEE Trans, on Fuzzy Systems, 1998, 6(1) : 12-32． .
[24] Sulzberger S. M., Tscbicholg-Gurman N. N. and Vestli S. J. FUN: optimization of fuzzy rule based systems using neural networks. Proceedings of IEEE Conference on Neural Networks, San Francisco, 1993: 312-316．
[25] Kasabov N. Evolving fuzzy neural networks-algorithms, applications and biological motivation. Methodologies for the Conception. Design and Application of Soft Computing, World Scientific, 1998:271-274．
[26] Kasabov N. and Qun S. Dynamic evolving fuzzy neural networks with 'm-out-of-n' active nodes for online adaptive systems. Technical Report Tr99/04, Department of information science, University of Otago, 1999．
[27] Dasgupta D. Artificial Immune Systems and Their Applications. Spring-Verlag, 1999．
[28]Hunt J. E., Cooke D. E. Learning using an artificial immune system. Journal of Network and Computer Applications, 1996, 19:189-212.
[29]Tazawa I., Koakutsu S. and Hirata H. An evolutionary optimization based on the immune system and its application to the vlsi floor-plan problem. Trans of the Institute of Electrical Engineers of Japan (Series C and in Japanese), 1997, 117(7): 821-828.
[30]Tazawa I., Koakutsu S. and Hirata H. An optimization method based on the immune system. Proceedings of World Congress on Neural Networks, International Neural Network Society 1996 Annual Meeting, San Diego, CA, USA, 1996: 1045-1049.
[31]王磊，潘进，焦李成．基于免疫策略的进化算法．自然科学进展，2000，10(5)：451-455．
[32]王磊．免疫进化计算理论及应用．西安电子科技大学博士学位论文，2001．
[33]周家驹，何险峰译．演化程序—遗传算法和数据编码的结合．北京：科学出版社，2000．
[34]Kitana H. Empirical studies on the speed of convergence of the neural network training by genetic algorithm. Proceedings of AAAI-90, 1990.
[35]Whitley D., Starkweather T. and Bogart C. Genetic algorithms and neural networks: optimizing connections and connectivity. Parallel Computation, 1990, 14(3): 347-361.
[36]Whitley D. The GENITOR algorithm and selective pressure: why rank-based allocation of reproductive trails is best. Proceedings of 3rd International Conferernce of Genetic Algorithms and Their Applications, San Mateo, CA: Morgan Kaufmnn, 1989:116-121.
[37]Caudell T. P. and Dolan C. P. Parametric connectivity: training of constrained networks using genetic algorithms. Proceedings of 3rd International Conference of Genetic Algorithms and Their Applications, San Mateo, CA: Morgan Kaufmnn, 1989:370-374.
[38]Srinivas M. and Patnalk L. M. Learning neural network weights using genetic algorithms--improving performance by search-space reduction. Proceedings of IJCNN'91, Singapore, Vol. 3, 1991: 2331-2336.
[39]Garis H. GenNets: genetically programmed neural networks—using the genetic algorithm to train neural nets whose inputs and/or outputs vary in time. Proceedings of IJCNN'91, Singapore, Vol.2, 1991: 1391-1396.
[40]Janson D. J. and Frenzel J. F. Application of genetic algorithm to the training of higher order networks. Journal of System Engering, 1992, 2:272-276.
[41]Janson D. J. and Frenzel J. F. Training product unit neural networks with genetic algorithms. IEEE Expertise, 1993, 8:26-33.
[42] Montana D. and Davis L. Training feedforward neural networks using genetic algorithms Proceedings of 11th International Joint Conference on Artificial Intelligence. San Mateo, CA: Morgan Kaufmann, 1989:762-767．
[43] Bartlett P., and Downs T. Training a neural network with a genetic algorithm. Technical Report, Department of Electric Engineering, University of Queensland, Australia, 1990．
[44] Menczer F. and Parisi D. Evidence of hyperplanes in the genetic learning of neural networks. Biological Cybernetics, 1992, 66: 283-298．
[45] Skinner A. J. and Broughton J. Q. Neural networks in computational materials science: training algorithms. Modeling and Simulation in Materials Science Engineering, 1995, 3(3) : 371-391．
[46] Fogel D. B., Fogel L. J. and Parto V. W. Evolving neural networks. Biological Cybernetics, 1990,63(6) : 487-493．
[47] Greenwood G. W. Training partially recurrent neural networks using evolutionary strategies. IEEE Trans, on Speech Audio Processing, 1997, 5(1) : 192-194．
[48] Saravanan N. and Fogel D. B. Evolving neural control systems. IEEE Expertise, 1996, 10:23-27．
[49] Fogel D. B., Wasson E. C. and Porto V. W. A step toward computer-assisted mammography using evolutionary programming and neural networks. Cancer Letter, 1997,118(1) : 93-105．
[50] Stork D. G. et al. Preadaptation in neural circuits. Proceedings of International Joint Conference on Neural Networks, vol. 1, Washington DC, 1990:202-205．
[51] White D. and Ligomenides P. GANNet: a genetic algorithm for optimizing topology and weights in neural network design. IWAANN'93, Berlin: Springer-Verlag, 1993:322-327．
[52] Liu Y. and Yao X. Evolutionary design of artificial neural networks with different nodes. Proceedings of ICEC'96: Nagoya, Japan, 1996, 670-675．
[53] Oliker S., Furst M. and Maimon O. A distributed genetic algorithm for neural network design and training. Complex Systems, 1992, 6(5) : 459-477．
[54] Marin F. J. and Sandoval F. Genetic synthesis of discrete-time recurrent neural network. IWANN'93, vol. 686, Berlin: Springer-Verlag, 1993:179-184．
[55] Alba E et al. Fully automatic.ANN design: a genetic approach. IWANN'93, vol. 686, Berlin: Springer-Verlag, 1993:399-404．
[56] Fang J. and Xi Y. Neural network design based on evolutionary programming. Artificial Intelligent Engineering, 1997, 11(2) : 155-161．
[57] Maniezzo V. Genetic evolution of the topology and weight distribution of neural networks. IEEE Trans, on Neural Networks, 1994, 5(1) : 39-53．
[58] Yao X., Liu Y. A new evolutionary system for evolving artificial neural networks. IEEE Trans, on Neural Networks. 1997, 8 (3) : 694-713．
[59] Yao X. and Liu Y. Towards designing artificial neural networks by evolution. App. Math. Computation, 1998,91(1) : 54-65．
[60] Greenwood G. W. Training partially recurrent neural networks using evolutionary strategies. IEEE Trans, on Speech Audio Processing. 1997, 5(1) : 192-194．
[61] Kitano H. Designing neural networks using genetic algorithms with graph generation system. Complex Systems, 1990,4(4) : 461-476．
[62] Merrill J. W. J. and Port R. F. Fractally configured neural networks. Neural Networks, 1991,4(1) : 53-60．
[63] Narayanan A. and Moore M. Quantum-inspired genetic algorithm. Proceedings of IEEE International Conference on Evolutionary Computation, 1999: 61-66．
[64] Han K. H. and Kim J. H. Genetic quantum algorithm and its application to combinatorial optimization problem. Proceedings of the 2000 IEEE Congress on Evolutionary Computation, 2000: 1354-1360．
[65] Han K. H., Park K. H., Lee C. H. and Kim J. H. Parallel Quantum-inspired genetic algorithm for combinatorial optimization problem. Proceedings of the 2001 IEEE Congress on Evolutionary Computation, 2001: 1442-1429．
[66] Cortes C. and Vapnik V. N. Support vector networks. Machine Learning, 1995,20:273-297．
[67] Williams C. K. I. Computation with infinite neural networks. Neural Computation, 1998, 10(5) : 1203-1216．
[68] Bennett K. P. and Blue J. A. A support vector machine approach to decision trees. Proceedings of DCNN'98, Anchorage, Alaska, 1997: 2396-2401．
[69] Bennett K. Combining support vector and mathematical programming methods for induction. Advances in Kernel Methods-SV Learning, Cambridge, MA, MIT Press, 1999:307-326．