基于小波-支持向量机的工业取水异常数据挖掘与重构  

MINING AND RECONSTRUCTION OF INDUSTRIAL WATER ABNORMAL DATA BASED ON WAVELET-SUPPORT VECTOR MACHINE

作　　者：王晗 张峰 薛惠锋[1] Wang Han;Zhang Feng;Xue Huifeng(China Academy of Aerospace System Scientific and Engineering,Beijing 100048,China;School of Management,Shandong University of Technology,Zibo 255012,Shandong,China)

机构地区：[1]中国航天系统科学与工程研究院,北京100048 [2]山东理工大学管理学院,山东淄博255012

出　　处：《计算机应用与软件》2021年第5期61-68,81,共9页Computer Applications and Software

基　　金：国家自然科学基金青年项目(71904108);国家自然科学基金项目(U1501253);广东省省级科技计划项目(2016B010127005)。

摘　　要：提高取水监测数据质量是水资源管理中的紧迫问题。以工业取水监测数据为样本,梳理其异常类别,按照“粗筛选-精识别-再重构”思路,提出基于分段式3σ准则与小波变换、Fourier函数相结合的多尺度工业取水监测异常数据识别方法。采用自适应惯性函数与粒子群优化的最小二乘支持向量机模型重构恢复异常数据。结果表明,分段式3σ准则对数据的粗处理效果较好,采用Fourier函数可有效降低数据小波变换中的信息损失,提高异常数据的识别精准度。采用惯性函数-粒子群优化的LSSVM模型可满足异常数据较高精度的重构恢复需求,其重构精度强于LSSVM、PSO-LSSVM和传统的曲线拟合方法。该方法可为提高水资源数据的决策支持能力提供良好的方法参考。Improving the quality of water intake monitoring data is an urgent issue in current water management.The industrial water intake monitoring data was taken as a sample,the common abnormal categories of which were summarized,and the strategy of“rough screening-fine identification-reconstruction”was proposed.The multi-scale industrial water monitoring abnormal data identification models were constructed based on segmented criterion,wavelet transform and Fourier function.Moreover,the least squares support vector machine(LSSVM)model with adaptive inertia function and particle swarm optimization(PSO)was used to reconstruct the recovered anomaly data.Results show that the rough processing of the water monitoring data is better by segmented criterion,and the loss of data information in wavelet transform could be reduced by Fourier function,and then the abnormal data would be identified more accurately.The LSSVM model optimized by adaptive inertia function and PSO could accurately reconstruct and recover abnormal data,and its reconstruction accuracy is higher than LSSVM,PSO-LSSVM and traditional curve fitting method.Hence,this method could provide a good methodological reference for improving the decision support capabilities of water data.

关 键 词：取水监测 小波分解 多尺度挖掘 数据重构 

分 类 号：N945.2[自然科学总论—系统科学] TP399[自动化与计算机技术—计算机应用技术]