【作者】 夏天；

【导师】 管在林； 黄金国；

【作者基本信息】 华中科技大学 ， 机械工程， 2024， 硕士

【摘要】 物联网是生产基础设施的重要组成部分,物联网监测数据来源于对生产环境、设备和产品的实时监测,通过对此类数据的高质量采集和分析可保证生产质量。本文针对物联网监测环境,围绕时间序列数据异常检测与处理展开研究,主要的研究内容如下:首先,分析物联网监测时间序列数据异常检测与处理的现状与不足,完成了物联网监测时间序列数据异常检测与处理总体设计。总体设计包括异常检测与处理的需求分析、总体框架构建与相关算法理论的阐述。其次,设计并实现了物联网监测时间序列数据异常检测方法。该方法采用基于滑动窗口-改进差分自回归滑动平均模型-K-均值(SW-ARIMA-K-means)的算法。本方法流程包括:建立改进ARIMA模型的时间序列数据拟合与预测方法;建立了SWARIMA的异常检测方法;与K-means组合进行异常校验。通过实验结果验证了该检测方法的有效性与准确性。然后,设计并实现了物联网监测时间序列异常数据处理的组合修复方法。该方法以多项式插补方法、改进ARIMA预测方法和门控循环单元(Gated Recurrent Unit,GRU)预测方法等单一模型作为子模型,以决策树模型作为整合模型,对异常序列数据进行数据修复,提供参考修复序列。利用实例数据对该方法进行对比实验,验证了该方法的可行性与准确性。最后,设计了雪茄烟晾房监控数据异常检测及处理应用系统。该应用系统包括数据采集与可视化、异常检测、异常数据处理等模块。通过对系统的实现和应用验证了其设计的可行性与实用性。更多还原

【Abstract】 The Internet of Things(Io T)is an important component of production infrastructure,and its monitoring data comes from real-time monitoring of the production environment,equipment,and products.High quality collection and analysis of such data can ensure production quality.This thesis focuses on the monitoring environment of the Io T and conducts research on anomaly detection and processing of time series data.The main research content is as follows:Firstly,the current situation and shortcomings of anomaly detection and processing in Io T monitoring time series data were analyzed,and the overall design of anomaly detection and processing in Io T monitoring time series data was completed.The overall design includes requirement analysis for anomaly detection and processing,framework construction,and explanation of relevant algorithm theories.Secondly,we designed and implemented an anomaly detection method for Io T monitoring time series data.This method adopts an algorithm based on SW-ARIMA-Kmeans.The process of this method includes: establishing a time series data fitting and prediction method for improving the ARIMA model;established an anomaly detection method for SW-ARIMA;perform anomaly verification with K-means combination.The effectiveness and accuracy of the detection method have been verified through experimental results.Then,a combined repair method for handling abnormal data in Io T monitoring time series was designed and implemented.The method uses single models such as polynomial interpolation method,improved ARIMA prediction method and GRU prediction method as submodels,and the decision tree model as the integration model to repair abnormal sequence data and provide reference repair sequence.Comparative experiments were conducted using instance data to verify the feasibility and accuracy of this method.Finally,a monitoring data anomaly detection and processing application system for cigar drying rooms was designed.This application system includes modules such as data collection and visualization,anomaly detection,and anomaly data processing.The feasibility and practicality of its design have been verified through the implementation and application of the system.更多还原