2 0
1
7年8月
农业机械学报
第48卷第8期
doi：10．6041／j．issn．1000—1298．2017．08．017
基于SW—SVR的畜禽养殖物联网异常数据实时检测方法
璐1
段青玲1’2
肖晓琰1
刘怡然1
(1．中国农业大学信息与电气工程学院，北京100083；2．北京市农业物联网工程技术研究中心，北京100097)
张
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
window and
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El与支持向量回归(Sliding
support
for
vector machines
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0．1884，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。
关键词：异常数据检测；畜禽养殖物联网；滑动窗口；支持向量回归
中图分类号：TP274+．2；TP393．03
文献标识码：A
文章编号：1000—1298(2017)08旬159—07
Data Real-time Detection Method of Livestock
Anomaly
Breeding
Internet of
on SW——SVR
Based
Things
DUAN
XIAO
LIU Yiranl ZHANG Lul
Qinglin91· 2
and Electrical
(1．College ofInformation
Xiaoyanl
Engineering，China
100083，China
Agricultural University，Belting
Research Center
Internet
2．Beijing Engineering
ofAgricultural
of Things，Bering 100097，China)
to bad work environment and network transmission
iS
to
abnormal
Abstract：Due
sensory
failure．it
easy
generate
quality
data in livestock
Internet of
order ensure the
to
of
things
sensory
breeding
characteristics
based on
system．In
to the
of
data flow
sensory
as
such
data，according
periodicity，temporality，infinity，etc．，
a
method was
window and
vector machines
detection in real
and
proposed
regression(SW—SVR)for
sliding
support
data
livestock
Internet of
abnormal
things sensory
breeding
time．Firstly，the sliding
of data flow from
sampling frequency
window size wasdecided
to the characteristic
the
according
Internet of
period
livestock
data within
window was selected as
sliding
breeding
value of
things system，and
history
in
model．Then．the sensor estimated measurement value at certain moment
the
input
prediction
Internet of
livestock
was
SVR
breeding
things system
predicted by using
model．Finally，the prediction
if
actual
the sensor
the abnormal
data was identif／ed
sensory
interval(PI)was calculated．and
of
measurement data fell out
abnormal
the PI．The abnormal data would be
the
data．The
Internet of
error value of with
prediction
replaced by
predictive
data detection method wastested
data flow from real 1ivestock
sensory
by
breeding
resuhs showed that the mean
absolute
things system．Experiment
percent
window size calculated
by
the
window method was
88 4．The correct detection rate of abnormal
had
0．1
sliding
data based on SVR model with radial basis function
with BP neural
kernel(RBF kernel)achieved 98％．which
data can be
network(BPNN)method．Abnormal
higher accuracy compared
effectively
detected and treated in livestock
Internet of
breeding
things system．
internet of
data
detection；livestock
Key words：anomaly
breeding
things；sliding window；support
vector machines for
regression
重要手段之一。在畜禽养殖物联网中，传感器按照
时间序列连续不断地采集温湿度、光照、有害气体浓
度等畜禽生长环境数据，并以数据流的形式传输至
引言
当前物联网已经成为畜禽养殖生产获取数据的
收稿日期：2016—12—14修回日期：2017一Ol一16
基金项目：国家高技术研究发展计划(863计划)项目(2013AAl02306)和山东省自主创新项目(2014XGAl3054)
作者简介：段青玲(1967一)，女，教授，博士，主要从事智能信息处理研究，E-mail：dqling@cau．edu．ca
万方数据
全部评论(0)