基于Bi-LSTM网络的管道异常数据检测方法  

Pipeline Anomaly Data Detection Method Based on CNN and Bi-LSTM Network

作　　者：李春生[1] 田梦晴 张可佳 LI Chun-sheng;TIAN Meng-qing;ZHANG Ke-jia(School of Computer and Information Technology,Northeast Petroleum University,Daqing 163319,China)

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163319

出　　处：《计算机技术与发展》2023年第6期215-220,共6页Computer Technology and Development

基　　金：国家自然科学基金项目(51774090);黑龙江省自然科学基金面上项目(F2015020);黑龙江省省属本科高校基本科研业务费东北石油大学引导性创新基金(2021YDL-12)。

摘　　要：在管道运行过程中,受技术、计量仪器和自然环境等影响,导致管道数据经常出现异常值,影响调度人员无法进行正确的决策,不利于管道监控系统的安全稳定运行。传统的时序数据异常检测方法的准确率和检测速度得不到保证。针对该问题,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)和双向长短期记忆(Bi-directional Long-Short Term Memory,Bi-LSTM)网络的管道异常数据检测方法。首先,研究管道异常数据的表征及异常数据的产生原因,对管道数据进行野点剔除、均值填充和归一化处理,后通过CNN对处理后的管道数据进行特征提取;其次,利用Bi-LSTM网络充分挖掘管道数据间的规律,训练得到预测模型;再次,确定动态阈值,通过计算预测值与真实值误差并与阈值进行比较,检测异常数据;最后,在真实应用场景测试,通过设计一系列对比实验验证了该方法在处理速度和检测准确率等方面具有明显优势,且检测异常点的准确率高于同类算法。In the process of pipeline operation,due to the influence of technology,measuring instruments and natural environment,pipeline data often appear abnormal values,which affects the scheduler cannot make correct decisions and is not conducive to the safe and stable operation of pipeline monitoring system.The accuracy and detection speed of traditional time series data anomaly detection methods are not guaranteed.Aiming at this problem,we propose a pipeline abnormal data detection method based on convolutional neural network and bidirectional long-short-term memory neural network.Firstly,the representation of abnormal pipeline data and the causes of abnormal data are studied,outfield elimination,mean filling and normalization are carried out on the pipeline data,and then the characteristics of pipeline data after processing are extracted by CNN.Secondly,Bi-LSTM network is used to fully mine the law between pipeline data and build a prediction model.Thirdly,determine the dynamic threshold,calculate the error between the predicted value and the true value and compare it with the threshold value to detect abnormal data.Finally,the proposed method is applied in a real world to demonstrate our ultra-short-term working condition prediction method that achieves superior results for prediction accuracy and running speed when compared with other methods,and the accuracy of detecting outliers is higher than that of similar algorithms.

关 键 词：异常点检测 管道运行数据 卷积神经网络 双向长短期记忆网络 时序数据 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]