摘要
为提高农业物联网的数据感知质量,提出了基于向量升维的异常农情数据实时检测方法。首先采用滑动窗口机制将标准化后的时序农情数据转换为观测向量,接着将相邻向量元素差值之和作为新向量元素对观测向量进行升维,最后构建了异常数据实时检测框架。采用畜禽养殖物联网环境数据进行实验,开展滑动窗口大小取值、分类模型的异常数据检测性能与敏感性分析。结果表明,滑动窗口大小取2为宜,提出的向量升维方法能够有效提升分类模型的异常数据检测能力,且线性核支持向量机具有较优的异常数据检测性能和计算耗时,其检测效果与数据波动性和采样间隔负相关、与异常值偏离幅度正相关。
To improve data sensing quality of agricultural Internet of Things,a real-time outlier detection method of agricultural data is proposed by means of increasing vector dimension.Firstly,time-series agricultural data are normalized and transformed into observation vectors using sliding window mechanism.Subsequently,the dimension increase of observation vector is implemented by summing the difference among neighboring vector elements.Finally,the real-time outlier detection framework of agricultural data is constructed based on vector dimension increase.The experiment is carried out,based on data from real livestock breeding Internet of Things,for evaluating sliding window size,outlier detection and sensitivity of classification models.The results show that the appropriate size of sliding window should be two,the proposed vector dimension increase can effectively improve the outlier detection ability of classification model,and support vector machine with linear kernel outperforms its counterparts in terms of outlier detection performance and computation overhead.Additionally,its detection ability is negatively correlated with data fluctuation and sampling interval,and positively correlated with outlier deviation.
作者
赵刚
饶元
王文
姜敏
江朝晖
ZHAO Gang;RAO Yuan;WANG Wen;JIANG Min;JIANG Zhaohui(School of Information and Computer Sciences,Anhui Agricultural University,Hefei 230036)
出处
《安徽农业大学学报》
CAS
CSCD
2021年第2期304-311,共8页
Journal of Anhui Agricultural University
基金
安徽省自然科学基金项目(1608085QF126)
安徽省重点研究和开发计划面上攻关项目(201904a06020056)
安徽农业大学省级大学生创新创业训练计划项目(201910364263)共同资助。
关键词
向量
升维
农情数据
异常检测
物联网
vector
dimension increase
agricultural data
outlier detection
Internet of Things
