融合特征距离与信息熵的大田土壤数据聚类方法

发布时间：2018-04-12 12:28

本文选题：无线传感器网络 + 传感器部署　；参考：《北京邮电大学》2017年硕士论文

【摘要】：农业是国家现代化的基础,也是全面建成小康社会的重点和难点。当前我国仍处于由传统农业向现代农业转变的关键时期。大田状态感知的数据量多、结构复杂,引入数据聚类方法能够有效挖掘感知数据的内在联系,从而为过滤冗余数据和合理优化传感器部署提供了可行方案。为此,本文在国家支撑计划的支持下,针对大田状态感知的数据冗余大和传感器部署混叠的问题,研究融合特征距离与信息熵的大田土壤数据聚类方法,主要研究内容如下:1、面向大田作业的数据效能分析和聚类算法支撑。研究当前大田作业的空气和土壤的特征变化情况,得到大田作业感知数据的维度和时效需求。研究蓝牙、射频、Zigbee等不同传感器网络在大田作业中的数据感知能力。分析BIRCH、STING、DENCLUE以及k-means等四种数据聚类算法的数据处理能力。2、融合特征距离与信息熵数据聚类算法。针对现有聚类算法的聚类簇难以确定,以及初始聚类中心敏感的问题,本文提出融合特征距离与信息熵数据聚类算法,对多维扩散的感知进行数据标准量化,降低原始数据的偏差值。基于欧式聚类因子对感知数据进行初级聚类,得到概略的聚类簇,限定数据的分类范围。基于概略聚类簇,仍需要进一步提升数据的聚类程度。针对非均匀稀疏数据的处理效果差的问题,提出融合熵增减的聚类优化,构建基于熵增减的多目标聚类准则函数,通过级联熵增减的多目标聚类收敛方法,实现面向大田作业状态感知数据的最优聚类。通过仿真表明,融合特征距离与信息熵数据聚类算法能够提升数据能效2.3%。3、研制特征距离与信息熵数据聚类的大田墒情监测与预测系统。设计并完成大田墒情监测与预测系统的总体架构、硬件架构、软件架构和数据库字典。基于级联熵增减的非均匀稀疏数据聚类算法,优化大田传感器部署。通过搭建大田墒情监测与预测系统对融合特征距离与信息熵的大田土壤数聚类方法进行测试和性能分析。利用河南长葛试验田采集到的数据为聚类样本进行聚类,根据聚类结果指导当地的传感器部署,大田覆盖信息度为93%,利用优化前后的数据对大田墒情进行预测。实测数据表明,基于融合特征距离与信息熵的数据聚类方法的传感器部署方案采集的数据对大田土壤墒情预测平均误差为0.016。
[Abstract]:Agriculture is the foundation of national modernization and the key point and difficulty of building a well-off society in an all-round way.At present, our country is still in the key period of transition from traditional agriculture to modern agriculture.Because of the large amount of data and complex structure, the data clustering method can effectively mine the internal relationship of the perceived data, which provides a feasible scheme for filtering redundant data and optimizing the sensor deployment.Therefore, in this paper, with the support of the State support Plan, aiming at the problem of large data redundancy in field state perception and the problem of sensor deployment aliasing, a field soil data clustering method based on the fusion of feature distance and information entropy is studied.The main research contents are as follows: 1, data efficiency analysis and clustering algorithm support for field operations.The changes of air and soil characteristics of field operations were studied, and the dimension and time requirement of field job perceptual data were obtained.The data sensing ability of different sensor networks such as Bluetooth and RF Zigbee in field operation is studied.This paper analyzes the data processing ability of four data clustering algorithms, such as Birch and STINGNCLUE and k-means, and combines the feature distance and information entropy data clustering algorithm.In view of the difficulty to determine the clustering clusters of the existing clustering algorithms and the sensitivity of the initial clustering centers, this paper proposes a data clustering algorithm based on the fusion of feature distance and information entropy, which quantifies the perception of multidimensional diffusion.Reduces the deviation value of the original data.Based on the Euclidean clustering factor, the perceptual data is preliminarily clustered, and a general clustering cluster is obtained, which limits the classification range of the data.Based on the general clustering, it is necessary to further improve the clustering degree of data.Aiming at the problem of poor processing effect of non-uniform sparse data, the clustering optimization of fusion entropy increase or decrease is proposed, and the multi-objective clustering criterion function based on entropy increase and subtraction is constructed, and the multi-objective clustering convergence method based on cascade entropy increase or decrease is proposed.The optimal clustering for field job state perception data is realized.The simulation results show that the feature distance and information entropy data clustering algorithm can improve the data energy efficiency of 2.33. 3. A field soil moisture monitoring and forecasting system based on feature distance and information entropy data clustering is developed.The overall structure, hardware structure, software architecture and database dictionary of the field moisture monitoring and forecasting system are designed and completed.A nonuniform sparse data clustering algorithm based on concatenated entropy increases and decreases to optimize sensor deployment in the field.The field soil moisture monitoring and forecasting system was set up to test and analyze the performance of the field soil number clustering method which combines the characteristic distance and the information entropy.The data collected from Changge experimental field in Henan Province were used as clustering samples. According to the clustering results, the local sensor deployment was guided. The information degree of field coverage was 933. The soil moisture content was predicted by the data before and after optimization.The measured data show that the average error of soil moisture prediction based on the sensor deployment scheme based on data clustering method based on feature distance and information entropy is 0.016.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S126;TP311.13

【相似文献】