Computational Intelligence And Its Applications In The Analysis Of Meteorological Information

The macroscopical analysis of human intelligence is mainly presented by the ability to conceptualize the world at different granules and translate from one abstraction level to the others easily. The theory of quotient space granular computing represents the attributes of an object as well as the relationship among objects, and at the same time, describes the transition, transformation, mergence and decomposition among different grain-size worlds, therefore, this theory is a powerful tool to describe the macroscopical analysis ability human intelligence.The microscopical learning of human intelligence is mainly presented by the ability to summarize rules by learning large amounts of things, which can be realized through learning models. Covering algorithm, as the basis of constructive machine learning method, is one of the models that accord, which can simulate the microscopical learning ability of human intelligence.This dissertation studies quotient space granular computing theory and covering algorithm, which are the two models simulating human intelligence in two different ways, aimed at establishing formalized model for certain intelligent behavior and re-displaying partial function of human intelligence. The combination of the both is applied in the yield forecast of the winter wheat in Anhui Province, which forms the forecast models at different grain-size worlds. Experiments illustrate the universality of models.The dissertation includes:1. The outstanding function of quotient structure in quotient space granular computing theory is analyzed.To construct a proper abstraction level for a problem solver, how to obtain quotient space from the aspects of domain, attribute and structure are discussed to form a concrete idea, which serve as the theoretical basis for covering algorithm based on granular computing in Chapter Three, and the forecast model of winter wheat in Anhui Province based on granular computing in Chapter Six. The study is emphasis on the important function of quotient structure in quotient space theory, which is also considered as one of the major differences between quotient spacea theory and rough set theory.2. The covering algorithm based on granular computing is put forward, and the covering algorithm is applied in clustering fields.The rudiment of covering algorithm, including domain covering algorithm and alternative covering algorithm, is briefly introduced in this dissertation. Covering algorithm based on granular computing is put forward with relation to mergence method and attribute-granulation method in quotient space theory. Moreover, covering algorithm is introduced into clustering field, and that covering algorithm is applicable to clustering as well as classification is also indicated.3. The principle of "minimum covering" in covering algorithm is put forward.The principle of "minimum covering" in covering algorithm is put forward from the epistemological points of view, and the properties and characteristics of the minimum covering are further explored. The principle of"minimum covering" on covering algorithm is put forward from the perspective of geometry to form the sufficient and necessary conditions of minimum covering, and accordingly, the minimum covering algorithm. The complexity of minimum covering algorithm is also discussed. At last the minimum covering is solved with a programming way.4. Probability Model of Covering Algorithm is given.The kernel function and the idea of global optimization are introduced into covering algorithm to form the probability model of covering algorithm. In this algorithm, the decision function of covering algorithm has been changed into kernel function, and then the optimization is processed from the point of maximum likelihood principle of the statistical model to realize the automatic computing of the parameter of kernel function, which offers a way to solve the problem of parameter selection of kernel function.5. Forecasting modcls in different graih-size worlds are formed, according to the information of local weather and the yield of winter wheat collected from observatories that have been established in some of the cities in Anhui Province.1) Issue description is constructed in original space. The annual information from each city of Anhui Province forms an element in domain, namely, sample, which comprises two parts- characteristic attributes and decision attribute.2) The general methods of weather yield forecast are analyzed in this dissertation, and the disadbantages of these methods are pointed up. The improvement, accordingly, is put forward to realize the function of the process module of decision making attribute in the model constructed in this dissertation. After the yield phases are divided, the gray model is adopted to calculate the relative weather yield, which serves as the decision attribute of the sample.3) The methods of domain-granulation and attribute-granulation in quotient space theory are adopted to form different forecast models at different grain-size worlds.a) The granulation of domain: Cities between Yangtze River and Huai Rivers are set as examples. The elements which are closely related in structure or function belong to the same classification called [X]_jianghuai, which forms a forecast model, setting areas between Yangtze River and Huai Rivers as the whole.b) The granulation of attribute: The characteristic attributes of the sample, such as sunlight, water and temperature, are granulated by time stages to construct the quotient set [f]_ten-days, [f]_month and [f]_mix, respectively.4) At last, covering algorithm is adopted to carry out rules on these samples collected at different granules, and then forecast the real amount of yield.5) Samples about different cities between Yangtze River and Huai River are also mixed for learning and the results are satisfactory, which demonstrate the model is universalThe innovations of this dissertation are as follows:1. The principle of"minimum covering" of covering algorithm is put forward from the epistemological points of view, and the properties and characteristics of the minimum covering are further explored from the perspective of geometry to form the sufficient and necessary conditions of minimum covering, and accordingly, the minimum covering algorithm. The estimation of iteration times of minimum covering algorithm is also discussed.2. Covering algorithm based on granular computing is put forward with relation to mergence method and attribute-granulation method in quotient space theory.3. The combination of both quotient space theory and covering algorithm is applied in the yield forecast of the winter wheat in Anhui Province to form yield forecast models at different grain-size worlds.