基于SW-SVR的畜禽养殖物联网异常数据实时检测方法

发布时间：2018-04-17 02:21

本文选题：异常数据检测 + 畜禽养殖物联网　；参考：《农业机械学报》2017年08期

【摘要】：畜禽养殖物联网由于工作环境恶劣、网络传输故障等因素容易产生异常感知数据,为保证数据质量,根据畜禽养殖物联网数据流周期性、时序性等特点,提出了一种基于滑动窗口与支持向量回归(Sliding window and support vector machines for regression,SW-SVR)的异常数据实时检测方法。首先根据畜禽物联网数据流特征周期以及采样频率确定滑动窗口尺寸;然后通过SVR模型预测畜禽养殖物联网数据流中某一时刻传感器测量值;最后计算预测区间,根据实际测量值是否落入该区间判断是否异常并对异常数据进行置换处理。采用畜禽养殖物联网环境数据进行试验,结果表明:所提滑动窗口计算方法得到的窗口尺寸预测的MAPE为0.188 4,畜禽养殖物联网异常数据检测率达98%,能够有效检测和处理畜禽养殖物联网数据流中的异常数据。
[Abstract]:In order to ensure the data quality, the livestock and poultry breeding Internet of things is easy to produce abnormal perceptual data due to the bad working environment, network transmission failure and other factors. According to the characteristics of the data flow periodicity and timing of livestock and poultry breeding Internet of things,A method of real-time detection of abnormal data based on sliding window and support vector regression (SVM) is proposed.First, the sliding window size is determined according to the characteristic period and sampling frequency of the data stream of the Livestock and Poultry Internet of things; then the measured value of the sensor at a certain time in the data stream of the Livestock and Poultry Internet of things is predicted by the SVR model. Finally, the prediction interval is calculated.According to whether the measured value falls into the interval, the anomaly is judged and the abnormal data is replaced.Experiments were carried out by using the environmental data of the Internet of things for livestock and poultry breeding,The results showed that the MAPE of window size predicted by the proposed sliding window calculation method was 0.188, and the detection rate of abnormal data of livestock and poultry Internet of things was 98. It could effectively detect and deal with abnormal data in the data flow of livestock and poultry breeding Internet of things.
【作者单位】：中国农业大学信息与电气工程学院;北京市农业物联网工程技术研究中心;
【基金】：国家高技术研究发展计划(863计划)项目(2013AA102306) 山东省自主创新项目(2014XGA13054)
【分类号】：S818.9;TN929.5;TP391.44

【相似文献】