The Research Of Nature-inspired Computation Theory And Its Application In System Identification

Evolution over billions of years brings a diversity of lives to the nature, which is abundant of information processing mechanism. The purpose of nature-inspired computation is to investigate the function, characteristics, and mechanism of natural phenomenona, especially those of life, and build the corresponding computation models in order to serve the mankind society.A great deal of original data are provided in the problem of system identification and some mathematical models are abstracted from those data, with some rules concealed in those data found out. These problems sometimes are hard to be described through traditional certain models or solved with accurate mathematic methods, so they are more appropriate to be described as fuzzy optimization problems.Essentionally, fuzzy optimization methods are a kind of multi-object optimization problems, and genetic algorithms are a kind of effective means to solve multi-object optimization problems. Genetic algorithms in fuzzy optimization, especially those based on dynamic models, have their special properties, thus deep researches on them being essential.In the basis of expounding the theory and methods of nature-inspired computation, this dissertation focuses on two typical computation models of nature-inspired computation:fuzzy optimization and genetic algorithms. The fuzzy optimization based on genetic algorithms is researched by means of the thoery and methods of nature-inspired computation. About the problems on system identification, its object function is described by means of fuzzy optimization models which are solved through genetic algorithms and the validity, stability, fastness and high precision of the algorithms are demonstrated theortically and exemplified.The main researching work in the paper is as follows.(1)The basis concept, basis characteristics and main research areas of nature-inspired computation are expounded, the method to map the characteristics of the nature into computation models is researched, and the computation algorithms of nature-inspired computation systems are constructed. The coevolution computation is understood by means of the theory and methods of nature-inspired computation.(2)The basic concept of system identification is discussed and the classic and modern methods of system identification are analysed. The whole frame and identification mechanism of nature-inspired computation systems are studied, and the two typical frames, RFG and NFG, are constructed in the identification system. The characteristics of the source, scope of application, structure identification, parameter identification, signification, and so on of the two frames are analysed.(3) The basis properties on fuzzy optimization are expounded, and the traditional fuzzy optimization problems is discussed. The structural and parameter identification to regression equation and neural networks is researched, the utility functions(object functions) on parameter identification to regression equation and neural netwoke are constructed, and the fuzzy optimizition model is established to realize parameter identification by means of the maximum utility method.(4)The solution to fuzzy optimizition problems based on genetic algorithms is deeply inquired into. The methods of birthing initial group of genetic algorithms based on orthogonal design are introduced, the methods of adaptive genetic parameter adjustment are designed, the methods of the mutation along the weighted gradient direction and the adaptive adjustment of its weighted value are constructured, and other modified tactics of genetic algorithms is put out. The validity of the fuzzy optimization based on genetic algorithms is demonstrated theoretically.(5)Performance evaluation criteria for online and offline are designed and the measure of performance assessment in system identification are established. The problem about system identification based on RFG is described. The basic plan of the identification algorithm of regression models is explained and the main steps and means of the algorithm are constructed. The application of the algorithm to the design criterion identification of flash sizes is carried out; The problem about system identification based on NFG is described. The basic plan of the identification algorithm of neural netwok model is explained and the main steps and means of the algorithm are constructed. The application of the algorithm to the design criterion identification of flash metal consumption is carried out. The two identification algorithms are respectively analysed and compared, and the validity of the two algorithms is showed through some examples.The two system identification methods based on RGF and NFG are creatively put forword, and the self-organization ability, data parallelism, generalization ability, global optimization and adaptive ability of the two methods are demonstrated theortically and exemplified. With the valuable probe into the new system identification methods, the wide application prospect of the research will be showed.