基于近邻传播算法的茶园土壤墒情传感器布局优化

针对节水灌溉的土壤墒情传感器布局问题,提出了基于近邻传播算法(affinity propagation,AP聚类算法)的优化布局策略。策略在保证茶园传感网络全覆盖的基础上,实时采集试验区各节点的土壤墒情数据,构建节点土壤含水率的相似度矩阵,并迭代计算各节点的吸引度和归属度值,得出聚类数和聚类中心。结果表明,采用AP聚类算法对试验区域传感器进行优化布局,优化了传感器数量和位置,使传感器数量由25个减少到2个。在试验区随机采集土壤相对含水率,经验证,聚类中心节点的土壤相对含水率与试验区平均值相近,相对偏差近为0.76%,表明聚类中心节点的土壤墒情数据具有代表性。该方法有效降低了数据的冗余度,节约了系统成本。

Layout optimization of soil moisture sensor in tea plantation based on affinity propagation clustering algorithm

Abstract: Aiming at the layout problem of soil moisture sensors for water-saving irrigation, we proposed an optimal layout strategy of soil moisture sensors based on affinity propagation (AP) clustering algorithm. The soil moisture of tea plantation was as the research object. The tea plantation had 84-m width and 190-m length. Following the conventional method, 25 sensor nodes were evenly arranged in rectangular mode in tea plantation experimental area in order to guarantee full coverage of tea plantation sensor network. Soil moisture data of each sensor node in the test area was collected in real time for 3 days.The optimization of sensors was conducted based on soil water content and relative water content by AP clustering algorithm.Different clustering parameters were selected. The AP clustering algorithm was used to construct similarity matrix of node soil water content, to iteratively calculate the responsibility and availability of each node, and to form the clustering number and clustering center. When the clustering parameters were 10, 15, 20 and 25 times of preference, the AP clustering algorithm was used to calculate the soil moisture data in the experimental area for 3 days, the stable and consistent clustering results were obtained. Results showed that soil water content in the tested plantation presented an increasing trend from southwest to northeast and the largest difference of relative water content was 15%. The change is related to the topography of the tested area. For AP clustering, the maximum iterative times was designed as 1 000. Based on the results, the clustering result in the 3 days was 2. The number of sensors optimized by AP clustering algorithm was reduced from 25 to 2. The class mean of the relative water content of the soil in the experimental area was calculated, and compared with the relative water content of soil in the collection points of the cluster center, and the relative bias between them was less than 5%. The relative water content of the collection points in the cluster center was close to the average value of the experimental area, which indicated that the data collected by the cluster center can represent soil moisture situation in the experimental tea plantation. In order to verify the validity of this method, soil moisture data were collected randomly at 13 locations in the experimental area on January 2019. Results showed that the soil average relative water content of tea plantation in the experimental area based on 13 sampling points was 32.7%, the relative water content of soil in the cluster center based on 2 sensors was respectively 27.9% and 37% with an average in the cluster center of 32.45%. Compared with the average relative moisture in the experimental area, the relative bias was only 0.76%. It means that the AP clustering algorithm can optimize the distribution of soil moisture sensors in the experimental tea plantation. The relative soil moisture collected by the cluster center could reflect the overall situation of soil moisture in the tea plantation as long as using only 2 sensors arranged in the cluster center node determined by the optimization calculation. Thus, the AP clustering algorithm is suggested to use in optimization of the sensor layout, which can reduce the redundancy of data and accordingly realize cost saving in agricultural production system.