基于支持向量机的土壤基础肥力评价和土壤有机质含量预测研究

发布时间：2018-03-10 04:19

本文选题：支持向量机　切入点：等级分类　出处：《南京农业大学》2015年硕士论文　论文类型：学位论文

【摘要】：耕地土壤质量与肥力受到越来越多的关注,土壤有机质(Soil Organic Matter,SOM)作为土壤重要的养分来源之一,也成为研究的热点。土壤有机质含量预测是根据长期定位实验点的土壤有机质的含量变化进行预测,为研究土壤有机质提供科学理论依据。土壤养分评价是耕地土壤质量与土壤肥力评价的重要组成部分,常用的土壤养分指标为:土壤有机质,土壤全氮(Soil TotalNitrogen,TN),土壤全磷(Soil Total Phosphorus,TP),土壤全钾(Soil Total Potassium,TK),土壤速效氮(Soil Available Nitrogen,AN),土壤速效磷(Soil Available Phosphorus,AP),土壤速效钾(Soil Available Potassium,AK)的含量。在各种养分含量的基础上,分析耕地土壤质量与肥力的等级,可以为研究者提供科学合理的开发和管理土地资源的根据。目前数学建模在非数学领域也应用广泛,其中模型评价是数学建模的应用之一,例如综合指数法、人工神经网络法、模糊数学法、不同距离聚类法都是被大家广泛应用的方法。然而,这些模型不适合复杂非线性关系的因素的评价和土壤肥力水平的表现,需要在评估的过程中调整权重,影响了评价模型的覆盖度和结果的可靠性。近年来发展起来的支持向量机(Support Vector Machine,SVM)技术是在数学统计学习理论基础上发展起来的一种新型的机器学习技术,为实现上述目标提供了有效方法。该技术从VC维理论和结构风险最小化准则(SRM)的角度出发,保证模型能达到全局最优,具有最大泛化能力和强大的推广能力,能应用和解决许多预测问题,已成为机器学习领域颇有影响的成果之一。该研究在熟知支持向量机理论的前提下,与土壤生态相结合,做到理论与实际相结合:1、应用支持向量机分类理论评价湖南祁阳不施肥下红壤基础肥力等级。分析不同处理的核函数类型的支持向量机下土壤化学性质的实验数据。分析结果表明,支持向量机理论用于土壤基础肥力等级评价是可行的,并且还表明核函数类型对土壤基础肥力的类别不起决定作用。比较支持向量机模型分类结果与其另外三种评价方法(BP神经网络模型,判别法,聚类分析法)的分类结果,表明用支持向量机分类模型进行土壤基础肥力评价的结果与实测分类结果更可靠。2、应用支持向量机回归理论预测安徽阜阳土壤有机质的含量变化。实验数据通过支持向量机回归理论的方法与反向传播(BP)神经网络和径向基函数(RBF)神经网络相对照分析得出支持向量机模型结果更加精确。对土壤有机质含量和产量进行回归模拟,结果表明土壤有机质和作物产量呈正相关关系。3、为提高实验设计的完善性和数据的可靠性,本文提出了一种新型支持向量机预测和分类的土壤肥力分级模型一多重混合支持向量机模型。本研究将新型的机器学习方法-支持向量机方法应用于土壤生态领域,进行土壤基础肥力评价和土壤有机质含量预测,突出显示支持向量机方法的可行性和优越性。
[Abstract]:Soil quality and soil fertility of cultivated land has attracted more and more attention, the soil organic matter (Soil Organic, Matter, SOM) as one of the important sources of soil nutrients, has become a research hotspot. The prediction of soil organic matter content is predicted according to the change of soil organic matter content of the long-term experiment points, to provide scientific theoretical basis for the study of soil organic quality. Soil nutrient evaluation is an important part of soil quality and soil fertility evaluation of cultivated land, soil nutrient indexes commonly used for soil organic matter, soil total nitrogen (Soil, TotalNitrogen, TN), soil total phosphorus (Soil Total, Phosphorus, TP), soil total potassium (Soil Total Potassium, TK), soil available nitrogen (Soil Available Nitrogen, AN), soil available phosphorus (Soil Available, Phosphorus, AP), soil available potassium (Soil Available Potassium, AK) content. Based on the nutrient content of the soil, analysis of cultivated land Soil quality and soil fertility level, can provide scientific and reasonable development and management of land resources for researchers at present. According to the mathematical modeling in non mathematics fields are also widely used. The evaluation model is one of the mathematical modeling of the application, such as comprehensive index method, artificial neural network method, fuzzy mathematics method, clustering method is different from we are the widely used methods. However, these factors evaluation model is not suitable for complex nonlinear relationship and the soil fertility level, need to adjust the weights in the assessment process, affecting the reliability of the evaluation model and coverage results. Support vector machine developed in recent years (Support Vector Machine, SVM) technology is the development of learning theory in mathematical statistics. It is a new machine learning technique, provides an effective method to achieve the above goals. The technology from the VC dimension theory and knot Structural risk minimization (SRM) point of view, the model can achieve global optimal, with maximum generalization ability and strong generalization ability, and can be used to solve many problems of prediction, has become one of the most influential achievements in the field of machine learning. The research premise to vector machine theory as well as support, combined with soil ecology, achieve the combination of theory and practice: 1, evaluation of red soil fertility grade based Hunan Qiyang fertilizer under the application of support vector machine classification theory. The analysis of the experimental data of soil chemical properties of kernel function of support vector machine for different types of treatment. The analysis results show that the support vector machine theory for soil fertility evaluation is feasible., and also show that the category type of kernel function to soil fertility does not play a decisive role. The model of support vector machine classification results and other three kinds of evaluation methods (BP The neural network model, discriminant analysis, cluster analysis) classification results, the evaluation result showed that the soil fertility by using SVM model results and the measured results are more reliable classification.2, using support vector machine to predict the change of soil organic matter content in Anhui Fuyang regression theory. The experimental data with the back propagation through the method of support vector machine regression theory (BP) neural network and radial basis function (RBF) neural network according to the analysis of support vector machine model results more accurate. The soil organic matter content and yield regression simulation, the results show that the soil organic matter and crop yield was positively related to.3, in order to improve the reliability of experimental design and data integrity in this paper, proposed a soil fertility grading model of a novel support vector machine prediction and classification of heavy hybrid support vector machine model. This research will be the new type of machine Support vector machine (SVM) is applied in the field of soil ecology to evaluate soil fertility and predict soil organic matter content, highlighting the feasibility and superiority of support vector machine.

【学位授予单位】：南京农业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：S158;S153.621;TP18

【参考文献】