基于决策树的温室环境调控规则设计及其应用研究

发布时间：2018-07-17 03:55

【摘要】：随着无线传感网技术的发展和数据挖掘技术的兴起,基于数据驱动的精准农业已经成为一种新的发展趋势。在温室培育领域中,数据挖掘的任务就是将领域专家多年的成功培育经验转化成可用于调控温室小气候、预测产量等的决策规则和经验知识库。本文针对现有温室调控系统大多只是简单地采集环境参数,其决策规则的实现方式也只是通过设置简单的阈值,并且缺乏长期评估和有效反馈机制的问题,提出了一种学习算法来生成细粒度的温室作物培育规则。相比较于利用专家的培育经验来调控温室气候的方式,本文的细粒度决策规则可以在时间维度上对作物的生长气候做到较精准的控制。首先,将诸如空气温度和空气湿度的空间信息连同时间信息相结合进行分析,这样可以在保证作物品质的同时缩短培育周期。然后,为了能生成上述的自适应决策规则,我们将一个无监督的合并聚类算法应用到温室传感数据上,从而创建带有最佳时空特性标签的数据样本用于分类。接着,通过决策树模型的训练和测试,生成相应的作物培育规则。实验结果表明,我们的细粒度决策规则有着近80%的分类准确率。我们方法的优势在于,温室培育人员在没有培育经验的情况下,也可以实现对温室环境的较精准控制,而且方法具有普适性,可以推广应用到多种温室作物上。最后在对上述方法理论研究的基础上,开发了一个综合传感、决策、执行反馈的温室花卉自适应调控系统。
[Abstract]:With the development of wireless sensor network technology and the rise of data mining technology, precision agriculture based on data drive has become a new development trend. In the field of greenhouse cultivation, the task of data mining is to transform the successful cultivation experience of field experts into decision-making rules and empirical knowledge base which can be used to regulate greenhouse microclimate and predict yield. In view of the fact that most of the existing greenhouse control systems simply collect environmental parameters, their decision rules are implemented only by setting simple thresholds, and lack of long-term evaluation and effective feedback mechanism. A learning algorithm is proposed to generate fine-grained rules for greenhouse crop cultivation. Compared with the way of using the experience of experts to control the greenhouse climate, the fine-grained decision rules in this paper can accurately control the growing climate of crops in time dimension. Firstly, spatial information such as air temperature and air humidity is combined with time information to ensure the quality of crops and shorten the breeding period. Then, in order to generate the adaptive decision rules mentioned above, we apply an unsupervised merging clustering algorithm to greenhouse sensor data, and create data samples with the best spatio-temporal characteristics tag for classification. Then, through the training and testing of decision tree model, the corresponding crop cultivation rules are generated. The experimental results show that our fine-grained decision rules have nearly 80% classification accuracy. The advantage of our method is that the greenhouse cultivator can also achieve more accurate control of greenhouse environment without training experience, and the method is universal and can be applied to many kinds of greenhouse crops. Finally, based on the theoretical research of the above methods, an adaptive control system for greenhouse flowers is developed, which integrates sensing, decision making and execution feedback.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S625;TP311.13

【相似文献】