农业传感网络数据差错分析算法研究

Research on Data Error Analysis Algorithm of Agricultural Sensor Network

农业气象观测要素精度的高低对观测效果有着重要影响,尤其是在野外复杂的环境下,影响温度变化的要素尤其多。目前大量用于野外温度观测的设备多为节点分散的无线传感网络,所用传感器为成本低、精度差的工业传感器,一旦出错,难以排查,且对后续分析工作影响巨大。为了降低出错无线传感器的排查难度,及时锁定出错数据,从农业传感网络数据的角度入手,对其进行差错分析,从而对野外的无线传感网络进行出错排查。通过对自动站的标准数据进行充分提取分析,发现以一年为周期,同一天不同地区之间的最高温度差成近似正态分布,类比自动站所采集的温度数据特性,通过对工业传感器采集的数据进行提取、平滑等分析,发现了该性质也适用于工业传感器,由此设计了一种数据差错分析算法,应用于野外的工业传感器的检错,通过温度这一农业气象观测的关键要素之一进行验证,验证结果证明了该算法的有效性。该检错方法不直接对现有无线传感网络进行操作,而是对其进行数据无量纲化处理,再对其进行提取、平滑和对比等处理来检错,节省了传感器检错成本,节省人力物力。

The accuracy of the agrometeorological observation elements has an important impact on the observation effect, especially in the complex environment of the field, especially the factors affecting the temperature change. At present, a large number of devices for field temperature observation are mostly wireless sensor networks with scattered nodes. The sensors used are low-cost, low-precision industrial sensors. Once they are in error, they are difficult to check and have a great impact on subsequent analysis. In order to mitigate the difficulty in searching the dysfunctional wireless sensors and targeting the problematic data, we analyze error from the perspective of agricultural sensor network data, and subsequently check out the errors in the wild wireless sensor network. By fully extracting and analyzing the standard data of the automatic station, it is found that the maximum temperature difference between different regions on the same day is approximately normal distribution on a one-year cycle, and the temperature data characteristics collected by the analog automatic station are collected by industrial sensors. The data were extracted and smoothed, and it is found that this property is also applicable to industrial sensors. Therefore we design a data error analysis algorithm, applied to wild industrial sensor calibration. By validating temperature, which is one of the essential factors in agricultural atmosphere observation, the result demonstrates the efficiency of the algorithm. Instead of directly manipulating on the current wire-less sensor network, the algorithm conducts dimensionless processing, followed by extraction, smoothing, and comparison to inspect error, to save the cost of error inspection.