【作者】 陈桂芬；

【导师】 刘大有；

【作者基本信息】 吉林大学 ， 计算机应用技术， 2009， 博士

【摘要】 随着“3S”技术在农业领域的不断普及,农业数据增长迅速,农业已成为空间数据挖掘最富有机遇与挑战性的应用领域之一。本文是在实施国家“863”项目“玉米精准作业系统研究与应用”的过程中,基于土壤肥力数据库和玉米精准作业的要求,利用空间数据挖掘技术,提出了解决玉米精准施肥、土壤肥力评价、地力等级分类和产量预测等问题的新方法,研究成果已成功应用于玉米精准作业智能决策系统中。主要工作和创新点:1.进行了基于空间模糊聚类算法的玉米精准施肥的研究。使用模糊聚类分析方法,建立土壤养分分类模型;利用八连通法进行空间聚类分析,并将模糊聚类结果应用于空间聚类。这种两阶段聚类方法优于传统的单阶段聚类,其分类结果对玉米精准施肥具有重要的指导意义。2.提出了基于加权的空间模糊动态聚类算法及在土壤肥力评价中的应用。该算法与基于模糊等价关系的传递闭包方法进行比较表明,其聚类准确率要明显高于未加权的模糊聚类算法。将其改进的算法运用到精准农业的土壤肥力评价中,与实际情况相符。3.研究了基于粗糙集-决策树的优化算法及在地力评价中的应用。研究结果表明基于聚类的样本优选方法去除了大量冗余样本,基于粗糙集的属性约简方法去除了部分冗余属性,使用决策树方法构建决策树,节省了时间和空间,降低了模型的复杂度。因而,本文提出的聚类和粗糙集约简相结合的方法在时间、空间和准确性方面均优于其他方法,该算法能有效提高土壤地力等级分类的准确性和客观性。4.采用时间序列算法中的滑动求和自回归方法(ARIMA)来对玉米产量进行预测,实验结果表明应用ARIMA模型预测的玉米产量与实际值拟合效果很好。5.设计并实现了玉米精准作业智能空间决策支持系统(MPISDSS)。该系统将具有空间信息处理功能的地理信息系统、具有空间信息分析功能的空间数据挖掘技术、人工智能领域中的专家系统技术与传统的信息管理系统、决策支持系统有效集成,并将GIS中的统计分析方法与数据可视化结合起来,极大提高了农业管理部门进行农业生产决策的能力。更多还原

【Abstract】 Spatial data mining is the emerging multi-disciplinary cross-edge disciplines with the development of computer technology database applications and management decision support technology. With the growing popularity of 3S technology in agricultural and the rapid growth of agricultural data agricultural Information Science has become one of the richest opportunities and challenges of applications of the spatial data mining. This article discusses several key issues of spatial data mining and intelligent spatial decision-making that corn precision operation intelligent decision support system involved the process of the implementation of the national 863 project of corn precise operating system research and applications and presents a space-based data mining of the corn precision operation intelligent spatial decision support system and from the theory and practice of the system to explore research and application and carries out exploratory research and application from the theory and practice. Based on requests of fertility database and corn precision operation using of spatial data mining this paper presents a new method to solve soil fertility corn precise fertilization soil fertility level and yield forecasting. The research result has been successfully applied to maize precise operating intelligent decision-making system. Major work and innovation points in the following areas:1. For the Spatial cluster research present situation of spatial clustering and the actual needs of precision agriculture intelligent decision-making in order to solve regional problems of corn precision fertilizer corn precision fertilization based on spatial fuzzy clustering algorithm has been researched. Using fuzzy clustering analysis method to study the classification of soil nutrient studies and establish classification model using 8-connected to carry on spatial clustering analysis and the results of fuzzy clustering applied to spatial clustering. This two-stage clustering method is superior to the traditional single-stage clustering. It has some significance for spatial data mining and theoretical studies. The main contribution of this approach is that :(1) Spatial fuzzy clustering method for the first time applies to soil nutrient content analysis; (2) The existing clustering algorithms are simple non-spatial attribute clustering or spatial location based on nearby clustering but for precision agriculture in the precise fertilization must also consider the attributes and spatial location combination of the two clustering algorithms have not been reported previously. This classification is in line with the needs of corn precise fertilization. The classification has an important guiding significance for corn precision fertilizer and other field management and operation of precision agriculture to do decision making so make precision agriculture toward a more practical direction.2. Spatial fuzzy clustering algorithm based on weighted has been proposed application of weighted spatial fuzzy dynamic clustering algorithm can be able to conduct more accurate evaluation of soil fertility. For variable rate fertilization on soil nutrient study on the impact common practice is to compare soil N P K and other nutrients and do not comprehensive consideration of various nutrients in the soil with each other. There is also a part of the study of variable rate fertilization effects on soil nutrients. This paper presents an improved spatial fuzzy dynamic clustering algorithm and its application in soil fertility evaluation in view of traditional spatial fuzzy clustering algorithm insufficiency.The specific process including: first obtains various attribute weight using the analytic hierarchy process then unifies the weight and the spatial fuzzy dynamic cluster law in the end use F distribution in probability statistics to determine the best classification to improve the spatial fuzzy clustering algorithm for intelligent. Compared the method of the paper with the transitive closure method based on fuzzy equivalence relation.The experimental results show that: the algorithm clustering accuracy rate is significantly higher than the weighted fuzzy clustering algorithm. The improved algorithm is applied to soil fertility evaluation of precision agriculture the experimental results is in line with the actual situation. It makes a contribution to the accurate evaluation of soil fertility.3. Research on the optimization algorithm based on rough set– decision tree and application on soil fertility evaluation. Looking from data mining’s angle the soil evaluation materially belongs to the classified forecast question. Decision tree method is a better classification method suitable for dealing with nonlinear data and classification and predicted decision-making problems of describe data the speed of production decision tree is quick easy to extract the rules easy to understand. When the attributes in data set are excessively more some situations is easy to be appear when classifies with the decision tree such as decision tree structure is not good discover some useful rules information is difficulty and so on.In the rough set theory knowledge is interpreted as a variety of division on the domain model using two precision approximation and lower approximation sets to approximate the collection to be described is suitable for dealing with uncertain knowledge. Rough set theory can support data mining and knowledge discovery in multiple steps such as data preprocessing data reduction rule generation data acquisition and other dependencies. So rough sets and decision tree have a strong advantage of complementarities. This article uses the clustering method to obtain samples and the samples is more typical checked by decision tree model so the samples can be used as a standard model of learning samples; using rough set attribute reduction method combined with decision tree method to evaluate the productivity grade of soil that is use the rough set method to reduce the land attributes and obtain a low-dimensional training data then use decision tree method to build decision tree obtain classified rule sets. This method not only reduces the tree’s branches removing redundant attributes improve mining efficiency; and also enhance the model accuracy. The algorithm can effectively improve the soil fertility level classification accuracy and objectivity. And provides new ideas and methods for spatial data mining and knowledge discovery.4. Use of time series algorithm to predict corn yield has been studied corn production forecast is an important part of the implementation process of precision agriculture the traditional statistical methods are relatively large errors the cost of production forecast application of remote sensing maps is relatively high. In this paper time series algorithm method of sliding summation autoregressive (ARIMA) is used to predict the yield of maize and makes up for the shortcomings of statistical methods big errors but also remotes sensing measurements to verify the results produced. Experimental results show that the predict corn yield application of ARIMA models is very good with the actual value fitting effect and indicate that time series algorithm can be used on maize production to better predict future trends. This has provided the new mentality and the method for the corn rate of prediction analysis.5. Corn precise assignment is a complex spatial decision-making tasks this article Uses artificial intelligence techniques spatial analysis methods statistical analysis methods digital maps image analysis and visualization methods and so on in-depth analysis space factors time factors that accurate operations variables affecting corn fertilizing spraying variables intelligent measuring production that impact on variable rate fertilization variable spraying intelligent measuring production in corn precision operation and design and implement the maize precise intelligent space decision-making system.The system will effectively integrate geographic information system with spatial information processing functions spatial data mining techniques with spatial information analysis capabilities expert system technology in the field of artificial intelligence traditional information management systems Decision Support System and combine the statistical analysis of GIS methods and data visualization thus enhancing the capacity that agricultural administrative department full use of various properties and spatial information data to make decisions.The actual application shows that the system is very suitable for precision agriculture intelligent decision-making and further enriched and perfected the practice-oriented field study of spatial data mining techniques and methods.更多还原