【作者】 金妙光；

【导师】 赵涛；

【作者基本信息】 天津大学 ， 管理科学与工程， 2009， 博士

【摘要】 精准农业是通过地理信息系统、全球卫星定位系统、遥感技术等技术的应用，按照具体条件，精细准确地调整土壤和作物的各项管理措施，最大限度地优化使用各项农业投入，以获取最高产量和最大经济效益，同时保护农业生态环境、保护土地等农业自然资源。精准农业包括精准种子工程、精准播种、精准施肥、精准灌溉、作物动态调控及精准收获等技术系统。所以精准农业是一个复杂的综合性系统，作为多系统交叉的大型系统工程是数据驱动的，在数据驱动中数据挖掘技术尤为关键，数据挖掘技术主要内容是数学模型的选择与建立。在精准农业这一领域专家系统和人工智能提供的数据分析系统为决策支持系统提供支持，作为决策的前提和依据，预测数据的精确程度直接影响决策的可行性，所以预测是精准农业决策支持系统的核心，目前我国精准农业的预测还是一项空白。在实际预测的工作中任何一个单一预测方法都不可能利用全部有用的信息，而不同的预测方法往往能提供不同的有用信息。将不同预测方法进行适当的组合得到的组合模型可以达到改善预测精度的目的。所以针对不同的研究目标，根据其研究性质的不同，灵活、有效的选取适当的数学模型才能够保证预测结果的准确、可靠，才能真正做到“精准”。在实际的分析过程中影响预测的因素有很多，因素之间相互影响、关联，所以我们一般遇到的预测问题都是多元的问题。传统的数学方法可以解决多元线性问题，也可以解决一些简单的多元非线性问题。对于多元非线性的问题，传统的数学方法可能会出现求解复杂、预测偏差大或无法预测等问题，导致无法预测。伴随基于神经网络这一新兴科学产生的数学方法，例如：BP神经网络、径向基函数神经网络、广义回归神经网络等的出现，为精准农业技术领域预测手段的改进和完善提供了新的方法。由于神经网络具有极强的容错能力，使之能够轻松实现复杂的非线性映射。由于有着大规模的计算能力，使之在计算机、自动化及人工智能都实现了广泛应用，解决了许多传统方法难以解决的问题。当然作为一门新学科的出现，神经网络提供的数学方法也有着自身的局限性和缺点，但是随着神经网络这一技术手段的日趋成熟，将给我们带来翻天覆地的变化。MATLAB工具箱有大量的运算工具、绘图工具，还有非常简洁的人机交互界面，为复杂系统的数值计算提供了保障。神经网络与MATLAB的有机结合可以对包含大量不确定因素的系统进行分析、计算，使复杂的预测问题得到解决。更多还原

【Abstract】 Precision agriculture, which can precisely adjust managements about soil andplants according to the practical circumstance by making use of geographicalinformation systems , global positioning system and remote sensing technology, hasbeen optimizing the application of agricultural investments and maximizes theagriculture output and economy benefit, it protects agricultural resources such asenvironment and soil as well as. Precision agriculture, being comprised of interactedsystems of precise seeds engineering, precise planting, precise fertilizing, preciseirrigating, flexible plants regulating and precise harvesting, is a complex system whichis driven by data and in the process the digging data is most important. The mainlytechnique of digging data technology is how to select and setup the mathematicmodel. In the precision agriculture, experts and artificial intelligence data analyzingsystems sustain the decision‐making system. Being the precondition and groundworkof making decision, the precision of speculating and forecasting data dominates thefeasibility of the decision. So the speculating and forecasting data is the focus in themaking decision and sustaining systems in the precision agriculture, but it is still ablank in our counties precision agriculture.None of the alone implement can use every available information in practicalforecasting data. It is always happened that different result will be given whendifferent method is taken. The better precision can be obtained by using appropriatecombination of several different speculating implements. Therefore, only whenproper mathematic model is selected flexibly and effectively according to differentproperties of different studying object, can the precise and reliable predicts be makesure. There are many factors which have an influence on the forecasting in thepractical analyzing process. The factors are interacted and associated with each other,so it is normal that the problems of forecasting always are concerned plural entities.With traditional math people can resolve problems of linear multifactor functionsand some simple problems of nonlinear multifactor functions, but for most of thenonlinear multifactor problems, the complicated or inaccurate forecasting resultswould be obtained when using traditional math tools, which make forecastunavailable. With the appearance of mathematic model base on artificial neuralnetworks, for example, BP neural networks, radial factors functions neural networks,broad return neural networks, it is possible to improve and optimize the forecasting and speculating data in the precision agriculture. For artificial neural networkspossess excellent ability of tolerating mistakes, it can carry out complicatednonlinear mapping easily. And because of its great ability of computing, it has beenbeing applied widely in computer, automation and artificial intelligence and by whichmany difficulties that can not be resolved by traditional implements have beenresolved. It is certainly that artificial neural network mathematic implement have itslimits and shortcomings for it is still a new coming thing, but with the growing of thetechnology, it will make great change for us.MATLAB, a large‐scale computer assistant software which possesses a largeamount of calculating and drawing tools in its toolbox and has a very compactpeople‐computer interface, ensures the complicated calculating being carried out.Integrating the ANN and MATLAB will make analyzing and calculating systemscontaining large amount of variables become possible, furthermore, the complicateddifficulty of forecasting will be resolved.更多还原