| 本文已被：浏览 1170次 下载 1099次 |
码上扫一扫！ |
| 基于改进K-medoids算法的土壤墒情传感器布局优化 |
|
汪涛1,张武1,2*,苗犇犇1,刘波1,王瑞卿1,张立付1,徐少翔1,饶元1,2,江朝晖1,2
|
|
|
| (1.安徽农业大学 信息与计算机学院, 合肥 230036;2.智慧农业技术与装备安徽省重点实验室, 合肥 230036) |
|
| 摘要: |
| 针对茶园土壤墒情传感器布局中传感器数量过多、数据冗余度过大的问题，采用改进的K-medoids方法，优化茶园土壤墒情传感器使用数量及部署位置。在保证茶园传感网络全覆盖的基础上，实时采集各传感节点数据；构造各传感器在不同天气条件下的时间序列数据，并运用三次样条插值法拟合成连续函数；应用谱排直法定义新的时间序列数据的距离函数替换K-medoids中的欧氏距离，将聚类中心作为最终的传感器布点；随机选取位置并采集土壤相对含水率以验证聚类中心作为传感器布点的代表性。采用改进前和改进后的K-medoids方法对2018-07—2018-08(试验Ⅰ)和2020-12(试验Ⅱ)采集的土壤墒情数据进行聚类。结果表明:1)改进的K-medoids将32个传感器减少到4个，改进前后簇中心墒情值与簇均值的相对偏差，试验Ⅰ由2.85%降到1.91%，试验Ⅱ由2.01%降到1.43%；2)改进K-medoids所得聚类中心相对含水率与试验区域平均值相近，相对偏差小于2%；3)以改进的K-medoids算法所得聚类中心作为起点的布点路径长度为82.4 m，使用8个传感器，优于改进前的106.5 m和10个传感器。基于改进K-medoids的布局方法能够优化传感器的数量和位置并且在不同天气条件下均适用。 |
| 关键词: 土壤墒情 传感器布局 相对含水率 时序数据 K-medoids |
| DOI：10.11841/j.issn.1007-4333.2022.08.20 |
| 投稿时间：2021-10-21 |
| 基金项目:安徽省重点研发计划-面上攻关项目(201904a06020056,202104a06020012)；智慧农业技术与装备安徽省重点实验室开放基金(APKLSATE2019X001)；2016年农业部农业物联网技术集成与应用重点实验室开放基金(2016KL05)；安徽高校自然科学研究重大项目(KJ2019ZD20) |
|
| Layout optimization of soil moisture sensor based on the improved K-medoids algorithm |
|
WANG Tao1,ZHANG Wu1,2*,MIAO Benben1,LIU Bo1,WANG Ruiqing1,ZHANG Lifu1,XU Shaoxiang1,RAO Yuan1,2,JIANG Zhaohui1,2
|
| (1.College of Information and Computer, Anhui Agricultural University, Hefei 230036, China;2.Anhui Key Laboratory of Intelligent Agricultural Technology and Equipment, Hefei 230036, China) |
| Abstract: |
| Aiming at the problems of too many sensors and too large data redundancy in the layout of soil moisture sensors in tea plantations, an improved K-medoids method was used to optimize the number and deployment location of soil moisture sensors in tea plantations. On the basis of ensuring full coverage of sensing network in tea plantations, the data of each sensing node were collected in real time. The time series data of each sensor under different weather conditions were constructed, and the cubic spline interpolation method was used to synthesize the continuous function. The Euclidean distance in K-medoids was replaced by the distance function of new time series data defined by profile alignment method, and the clustering centers were used as the final sensor placement. This study randomly selected the location and collected the soil relative moisture content to verify the representativeness of the cluster centers as the sensor distribution. Soil moisture data collected from 2018-07 to 2018-08(experiment I)and 2020-12(experiment II)were clustered using the pre-improved and post-improved K-medoids methods. The results show that: 1)The improved K-medoids reduces the number of sensors from 32 to 4. Comparing the relative bias between the soil moisture value in the centers of the cluster and the cluster mean value before and after the improvement, experiment I decreased from 2. 85% to 1. 91%, and experiment II decreased from 2. 01% to 1. 43%. 2)The relative moisture content of the cluster centers obtained by improved K-medoids are close to the average value of the test area, and the relative bias is less than 2%; 3)The clustering centers obtained by the improved K-medoids algorithm as the starting points of the distribution path length is 82. 4 m, using eight sensors, better than the original 106. 5 m and ten sensors. In conclusion, the layout method based on improved K-medoids can optimize the number and location of sensors and is applicable under different weather conditions. |
| Key words: soil moisture sensor layout relative moisture content time series data K-medoids |
