Application Of Rough Sets Theory In Knowledge Discovery Of Agriculture Decision Support System

Agricultural Decision Support System (Agricultural DSS) is a human-computer interaction system based on modern information technology and it is aimed at solving the problem of semi-structure decision in the agricultural field, helping the agricultural administrative staff, agricultural researchers and farmers to make correct decisions. There are some difficulties in the research field of knowledge discovery based on Agricultural DSS due to the following factors concerning agricultural production information: enormous quantity of data, complicated factors and extensively-covered information. Presently the research of knowledge discovery in agricultural field mainly involves Fuzzy Theory, Genetic algorithm, etc., which is accompanied by the bottleneck problem of knowledge acquisition in DSS. The reasons for this problem are that the application of these technologies needs datum experience information in the process of decision rules obtaining and reasoning and that the obtained results are not easy to evaluate and interpret.Rough Sets Theory (RS) was proposed by Z Pawlak in the early thel980s. It is a new mathematical tool used to deal with fuzzy and uncertain knowledge. It doesn't need experience information of the datum and is based on the concept of classification, the set approximate, similar classification and indiscernible relation. Rough Sets can lead to the decision of question or classification rules by knowledge reduction. It has wide application in the fields like machine learning, decision analyzing, course controlling, pattern recognition and data excavation and so on.Compared with the traditional method of uncertain datum processing, the most significant feature of Rough Sets Theory is that it doesn't need any experience information of data and therefore the uncertainty description of the question is comparatively objective. The present thesis discusses the knowledge expression and characteristic in mass information according to Rough Sets Theory, then products minimum decision rules by analyzing, reasoning and reducing. In addition, the thesis delves on the algorithm construction of knowledge reduction in decision table by using decision logic in Rough Sets Theory. Besides the thesis proposes the scheme of knowledge discovery of agricultural DSS on the basis of Rough Sets Theory.The research of this thesis is carried out according to the basic procedures of knowledge discovery: (1) Understanding the field knowledge and relevant experience information and defining the systematic goal; (2) Carrying out data pretreatment, including completing the incomplete data, determining the qualitative description of inaccurate data and the fuzzy and dispersion treatment of data; (3) Utilizing a certain data processing method to reduce the data, determine the useful characteristic parameter or variable of the system, abbreviate the system and set up corresponding mathematical model or logic rules; (4) Testing the mathematical model or logic rule excavated orobtained from the data, and explaining the result obtained model, and solving the objective problem with the excavated knowledge.At present, in view of the characteristic and task of above-mentioned knowledge discovery from the data, the widely used data-processing methods include RS, Fuzzy Theory, Artificial Neural Network, Genetic algorithm etc. As for dealing with the inaccurate, incomplete and uncertain data in knowledge discovery, Rough Sets Theory has such advantages as high accuracy, low calculation amount and the capability of eliminating redundancy. So the thesis mainly discusses data pretreatment, data reduction, obtaining of rules and incremental studying of system. The major research directions and findings are as following:First, in the thesis, the discretization problem in Rough Sets Theory is discussed. We analyze the characteristics of discretization and do classification research on the discretization algorithm from different angles. The thesis proposes an algorithm of discretization based on information entropy. The algorithm defines the condition attributes and the maximum dimension of each consecutive concept layer based on the experts' prior knowledge and acquires the dividing territory value of the maximum information gains through using Genetic Algorithm. This method can be used to deal with the dispersion of attribute value and the optimization of territory value in the process of agricultural data processing.Secondly, as knowledge reduction is the core content of Rough Sets Theory, this thesis mainly discusses the method of the attribute reduction of Rough Sets. Among the existed research results, the distinguishable matrix proposed by Skowron presents an outstanding clue for exploring the most ideal knowledge reduction. Since the distinguishable matrix-based attribute reduction method involves the process of searching the combination of attributes and transforming it into logical formula, it has the drawback of large amount of data and data calculation in the establishment of the example-drawing model. This thesis therefore proposes a new calculating method to improve the reduction speed of data attributes.Thirdly, considering the task of knowledge excavation in decision support system and the previously mentioned research achievements, the present study proposes a new knowledge discovery model of decision support system based on RS Theory. The thesis provides the calculation method for increment rule excavation, which brings about the increment learning model. When new examples are added, the whole system doesn't need the updating. It only needs the maintenance of certain index information to realize the increment induction learning. Thus this method has the characteristics of being more adaptable and more dynamic.Finally, the thesis proposes the RS method to solve the decision problem in the concrete agricultural support system. This method can be used to eliminate redundant attribute, reduce...