Research On Meteorological Data Mining Based On Quotient Space

Data mining is a course for mining latent, unknown but useful knowledge and information from practical data which is plentiful, incomplete, noisy, fuzzy and stochastic. Nowadays, the technologies of data mining are researched and applied to many fields including bank, telecom, insurance, traffic and so on. Classification and prediction are two kind of important forms for data analysis which can be used for constructing model that can describe the important data class or forecasting the data trend in future. Classification and prediction are adopted for crop yield forecast which is the research and application in the field of meteorological data mining. Crop yied forecast has important meaning for macro control police and influence in regional structure.For more effective meteorological data mining, this thesis introduces the quotient space granular computing theory,grey model,structural machine learning algorithm and so on. Granular computing theory is a new concept and paradigm for information processing. Just as a great umbrella, it covers all researches on theories, methodologies, techniques, and tools that make use of granules, which has become one of the main study stream in artifical intelligence. Nowadays, three main granular computing models are fuzzy set theory, rough set theory and quotient space theory. Quotient space theory is introduced by professor zhang ling and professor zhang bo. This theory uses granular view to analyze and describe the world. To analyze the world from different granular levels is a good way to recognize things more comprehensive and reasonable, and it can reduce the computational complexity. For example, the application in the field of heuristic search and path planning show its value for practicality. Grey system model is formed by accumulated temporal data series and filters probable stochastic data from original data. It can find connotative rule from temporal data series and gain less stochastic but more disciplinarian data series to mine inherence character. The main character of structural machine learning is that its net structure and parameter should be constructed by processing practical data. That is to say that the net structure and parameter is not prearranged but by processing data.The primary work of the thesis is listed below:(1) This thesis summarizes the development of granular computing at first. Then it lays stress on the introduction of the quotient space theory and the structural machine learning algorithm (covering algorithm).(2) The principle and realization of several models for meteorological data mining (crop yield forecast) is introduced. For the character of crop yield forecast, with the basic formula for yield per unit forecast, crop yield can be divided into trend yield and meteorological yield. The trend yield can reflect the factor of social productive force and the meteorological yield can reflect influence by meteorological factor. By the combination of the two kinds of yield, the crop forecast would be more reasonable and exact. So this thesis lays stress on the introduction of different compositive models.(3) For the shortage of the present models, this thesis introduces a new kind of model for meteorological data mining. Firstly, granular computing theory model which is introduced by professor zhang is adopted to analyze the meteorological data (sunlight, mean temperature, precipitation) from different granular layers. By the property and definition in quotient space computing model, the complex meteorological data from different granular layers are integrated, which can make the character of meteorological data more clear to satisfy the machine learning. For the trend yield forecast, this thesis adopts the grey model GM (1, 1). But for the meteorological yield forecast, this thesis uses structural machine learning (covering algorithm). By the combination of two algorithms, the veracity of prediction would be improved.(4) By the experiment of crop yield forecast for five districts in Anhui province, it can be analyzed that the veracity of the model based on quotient space theory is better than other models.The crop yield forecast in this thesis is an important field of meteorological data mining and it is a typical application of classification and prediction algorithms. It may be a new referrence for application in different fields with quotient space computing model. Its general ideas and detail design can be expanded to other application areas with the similar tasks of classifying, and it has a prosperous future.