【作者】 甄玉存；

【导师】 袁巧霞；

【作者基本信息】 华中农业大学 ， 设施园艺学， 2009， 硕士

【摘要】 温室是设施园艺中比较成熟的农业生产设施之一,它能够提供一个相对独立的环境系统,为作物的生长提供良好的环境条件,克服外界环境变化对作物生长的影响,而实现作物的周年性生产。为了更好地实现温室生产,对温室环境实施精确控制,越来越受到人们的关注。人工神经网络具有良好的自适应性、自组织及很强的学习、联想、容错和抗干扰能力,可灵活方便的对多因子的复杂未知系数进行建模。针对传统数学模拟方法不能良好解决的这些问题,本文利用神经网络方法,研究设计了一种BP神经网络来实现对3连栋塑料温室的环境建模,为温室环境的精确控制提供良好的前提条件。通过温室环境试验,研究不同条件下温室对温室内环境因子变化的影响,为温室环境控制提供理论依据。研究结果表明:温室处于不同的工况条件下,温室内的环境因子变化十分复杂,没有明显的规律可循,不可以用简单的数学方法对温室内的环境变化做精确的描述,各环境因子之间的关系复杂多变,具有复杂的交互作用,使温室数学模型建立较为困难。论文以华中农业大学工程技术学院温室试验基地为研究对象,进行4因素两水平正交试验获得试验数据,将温室外的环境参数作为输入参数,将需要控制的温室内的环境参数作为待预测的参数,建立BP神经网络数学模型,来实现参数预测。将原始数据进行预处理,并通过对不同网络结构、不同算法条件下建立的神经网络模型的综合比较,确定神经网络结构为9-11-5型,网络算法为traingdx在该条件下得到的预测效果较好,并能保证网络的良好的泛化能力。并提供了应用MATLAB神经网络工具箱编写的主要程序代码。为了验证该预测系统的准确性和实用性,论文选取80组温室内外环境试验数据对模型进行仿真和泛化处理,将利用该预测模型得到的结果和实际试验数据比较,表明预测效果良好,traingdx有很好的泛化能力。从而表明该预测系统有一定的使用价值,对华中地区温室生产有指导作用,对温室环境控制研究有参考的价值。更多还原

【Abstract】 Greenhouse horticulture is one of the most mature ways of agricultural production methods. It can provide relatively independent environmental systems and a good environment condition to overcome the impact of outside environment on crop growth, and to keep good annual crop production. In order to achieve better greenhouse production, to keep accurate control of the greenhouse environment has attracted more and more attention. Artificial neural network, which has good self-adaptability, self-organization, strong learning, fault-tolerance and anti-jamming capability; can be flexible on the complexity of multi-factor unknown coefficients model establishment. Because traditional methods of mathematical model can not solve these problems, the neural network methods is used to design BP neural network to establish 3 multi-span plastic greenhouse environmental model for its good prerequisite for accurately controlling the greenhouse environment.To provided a theoretical basis for the control of the greenhouse environment the impact of environmental factors on greenhouse was studied. The results indicated that the changes of environmental factors are complex under different working conditions in greenhouse. Owing to its no regularity, a simple mathematical method could not be used to accurately descript the various environmental factors in the greenhouse. The relationship between different environmental factors was complex and changeable, which was more difficult to establish the greenhouse mathematic model with interactions of environmental factors.The thesis took the plastic greenhouse in the Institute of engineering and technology of Huazhong Agricultural University as the research object, and obtained the test data from the 4 factors two-level orthogonal test. This test took the environmental parameters outside greenhouse as input parameters, and took the environment parameters inside greenhouse which needed for controlling as preparative parameters. These parameters were used to establish the forecast BP neural network model. The original data was pretreated, and the neural network model with different network structures and algorithms were comprehensive compared. The neural network had a better predictability and a good network generalization ability when the neural network structure was 9-11-5 and the network algorithm was traingdx. The main MATLAB neural network toolbox program code was provided in this paper. In order to verify the accuracy and practicality of the prediction system, the 80 group of environment test data outside and inside greenhouse were selectd to treat the model with simulations and the generalization, the results of the prediction model and actual test data were compared. It showed that the prediction result was good, and the traingdx had good generalization ability. The forecasting system has reference value for the environmental control of greenhouse in the middle of China.更多还原