| Data mining is a cross-disciplinary, influenced by a variety of disciplines, and therefore there are many different definitions. The more common: data mining is the discovery of large amounts of data from an interesting model, the data can be stored in databases, data warehouses or other information storage. Data Mining is an emerging cross-disciplinary, the technology has been successfully applied to the financial, retail, health care and government decision-making and other fields and achieved good social and economic benefits, there are broad prospects for the development and application.Clustering Data Mining is a very important part of the cluster analysis of the data set is divided into the data to a limited number of categories (or cluster called), made with a group of clusters of similar samples, which belong to different clusters of different samples A set of methods. In the fuzzy clustering algorithm for the analysis of the many, fuzzy c-means (abbreviated into FCM) can be said to be most widely used, the most sensitive of an algorithm. Fuzzy c-means clustering algorithm is the c-means clustering algorithm to the fuzzy form, each with a sample of each cluster from the center of its square-weighted membership, so as to the type of error in the goal square and function expansion For the type of error in the weighted sum of squares function.With the development of the information society, information-based agriculture and the rapid advancement of precision agriculture is receiving increasing attention. Precision agriculture in the world today is a new trend of agricultural development, information technology support is based on the spatial variability, location, timing, the quantitative implementation of a set of modern farming technology and operation management system, its basic meaning is based on the crop of soil properties, Adjustment of crop inputs, that is, on the one hand, to identify soil properties Field internal productivity and spatial variability, on the other hand, to determine crop production goals, locate the "diagnosis system, optimize the formulation, technical assembly, and scientific management" and the mobilization of soil productivity In order to save the most or the least input to achieve the same income or higher incomes, and improve the environment and efficient use of various types of agricultural resources, to achieve economic and environmental benefits.Precision application of precision agriculture is the core. Precision fertilization can be seen as a special kind of problem, at present the precise application of traditional strategy, quantitative methods are nutrient balance fertilizer and function of two [20]. Nutrient balance method more factor to be determined, and does not reflect the interaction between the nutrients are difficult to promote. It is generally believed to be accurate application of different spatial units of production data and other multi-layer data (of the physical nature of the soil, Diseases, Pests and Weeds, climate, etc.) based on analysis of the composite, with crop growth models, crop nutrition expert system for support to High-yield, high-quality, environmental protection and for the purpose of the variable fertilization prescription theory and technology. Precise application of information technology (RS, GIS, GPS), biotechnology, chemical and mechanical engineering technology optimization.In data mining technology used in a broader technology is the artificial neural network. ANN is a large number of parallel-processing unit composed of simple processing units, connected by adjusting the strength from the knowledge and experience to further study the capacity and knowledge can be applied. At present, the most widely used neural network model is the BP network, BP network is used Widrow-Hoff learning algorithm can be micro-and non-linear transfer function of the multi-layer network. Being spread, the role model in the input layer, hidden layer processing, the output layer to the mass. If the output level can not be expected output, then transferred to reverse the spread of the error stage, the output error by the adoption of some form of hidden mass to the input layer, layer by layer and return to the sharing of all units at all levels in order to obtain the The reference to the unit level error or error signal, as the right to modify the unit value basis. Such signals are spread and reverse the spread of error at all levels of the value of the right to modify the course of matrix, which is the network's learning process is to come and go. Neural network is integrated with a neural network is limited to the same issue of learning, neural network is integrated in the importation of a sample of the output constituted by the integration of the neural network in the sample under the co-decision output. Neural network ensembles include three elements: the individual generation of individual network synthesis methods and conclusions. To get and use of neural network integration, generally include: training to generate subset of neural network integration. To be used for training the individual networks. Bagging is a technology commonly used algorithms and Boosting algorithm.In this paper, the research work around the National Natural Science Foundation project, "the relationship between learning in a number of statistical research (60573073)" and Jilin Province Science and Technology Development sub-projects focus on the subject "digital space-time agricultural information management platform (20060213)" launched. In the study of neural network based on the application of clustering algorithm, BP algorithm, neural network technology integration, based on a cluster of selective neural network integrated approach, and applies it to solve the agricultural production of precision fertilizer problem. The article's integrated approach includes three main parts: the first use of Bagging a number of simple algorithm for BP neural network training; and then use the fuzzy c-means has been on the neural network to choose; final linear integrated approach to the selected nerve Network integration. The use of corn in 2007 Elm Three experimental field and on the 7th to the data of the experimental methods, experimental results show that the number of neural networks for linear integrated approach is not only well below the error of their individual neural network, and the generalization ability Relatively strong.In this paper, widely read in literature and in-depth understanding of the principles of data mining and application on the basis of the neural network integration, applications to do the following:(1) The standardization of data. 2007 Elm for experimental field of maize on the 3rd and on the 7th to the research and data aggregation.(2) Neural network training set. Bagging algorithm using a number of training BP neural network, thus ensuring the neural network to generate a collection of diversity; at the same time to retain only the goal is less than the threshold of the network, ensuring the effectiveness of a single individual networks. Test samples used in playback of data and the number of training set of similar size. In the neural network training before the first set of training [-1,1] standardization, and then the linear transformation in the role of verification and test-set.(3) Neural network to choose. Based on selective neural network integrated thinking, clustering technology from the resulting neural network, choose a collection of high-precision, diversity and strong network of individuals, instead of all network integration. First of all, given the similarity between the neural network measurement formula to calculate the similarity matrix neural network; and then use their clustering clustering algorithm to find the type of each type of cluster centers selected collection of neural network for network integration To prepare.(4) Linear neural network integration. Will be elected by the linear neural network, such as the right to integrate the results with a single neural network to compare the results. |
PDF Full Text Request