基于动态融合LOF的城市污水处理过程数据清洗方法  

Data-cleaning method based on dynamic fusion LOF for municipal wastewater treatment process

作　　者：鲁树武 伍小龙 郑江[3] 何政[3] 顾剑 韩红桂 LU Shu-wu;WU Xiao-long;ZHENG Jiang;HE Zheng;GU Jian;HAN Hong-gui(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China;Beijing Key Laboratory of Computational Intelligence and Intelligent System,Beijing 100124,China;Beijing City Drainage Group Company Limited,Beijing 100124,China;Beijing Drainage Water Environment Development Company Limited,Beijing 100022,China)

机构地区：[1]北京工业大学信息学部,北京100124 [2]计算智能与智能系统北京市重点实验室,北京100124 [3]北京城市排水集团有限责任公司,北京100124 [4]北京北排水环境发展有限公司,北京100122

出　　处：《控制与决策》2022年第5期1231-1240,共10页Control and Decision

基　　金：国家自然科学基金重大项目(61890930-5,61622301);国家自然科学基金创新群体项目(62021003);国家重点研发计划项目(2018YFC1900800-5);北京高校卓越青年科学家项目(BJJWZYJH01201910005020)。

摘　　要：围绕城市污水处理过程数据存在连续噪声和缺失的问题,提出一种基于动态融合局部异常因子(dynamic fusion local outlier factor,DFLOF)的污水处理过程数据清洗方法.首先,设计一种基于滑动窗口的数据动态分段方法,通过计算每个子段数据的均值、最大值和峰值区间信息获得数据异常属性值;其次,建立一种基于DFLOF的数据可信度评价模型,利用基于动态融合局部异常因子算法评估数据的可信度,保证异常数据检测和剔除的准确率;最后,提出一种基于径向基函数(radial basis function,RBF)神经网络的数据补偿方法对缺失数据进行补偿,实现污水处理过程数据的清洗.将该数据清洗方法应用于实际污水处理过程,实验结果表明:基于动态融合局部异常因子的数据清洗方法能够实现污水处理过程中异常数据的清洗,从而提高数据质量.In order to reduce the impact of continuous data noise and loss,a dynamic fusion local outlier factor(DFLOF)method is proposed for data-cleaning of the municipal wastewater treatment process(WWTP).First,a data dynamic segmentation method based on sliding window is designed to obtain the abnormal attribute of each segment,including mean value,maximum value and peak interval.Then,a data reliability evaluation model based on the DFLOF is established to evaluate each data segment by using the dynamic fusion local outlier factor algorithm,which improves the accuracy of abnormal data detection and elimination.Finally,a data compensation method based on radial basis function neural network is proposed to compensate the missing data and further realize the data-cleaning of the WWTP.The proposed cleaning method is applied to a real WWTP,the experimental results show that the data-cleaning method based on the dynamic fusion local outlier factor is able to clear abnormal data and improve the data quality.

关 键 词：污水处理过程 数据清洗 动态融合LOF 径向基函数神经网络 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]