Research On Data Cleaning Of FCCU Blowdown Based On I Forest And BPNN

At the present stage,the environmental protection demands put forward higher requirements for the standard emission of FCCU in refining and chemical enterprises.In the era of big data,it is necessary for enterprises to improve monitoring methods,combine machine learning methods to further explore and play the role of historical data,and reduce or avoid the occurrence of environmental pollution events.In the process of data sampling,transmission and storage,it will be affected by various factors and produce outliers.To avoid the influence of outliers on subsequent data analysis and prediction,data cleaning is an essential and important part.In this thesis,the research status of data cleaning methods at home and abroad is analyzed,and the existing methods of anomaly detection and missing data patching are summarized,which lays a foundation for future research.Secondly,in the outlier detection phase,an improved i Forest algorithm is proposed according to the characteristics of the blowdown data of FCCU.The algorithm combines the split criterion with relative mass to optimize the branching step of the i Forest,which improves the accuracy of the model and verifies the performance of the algorithm through experiments.Finally,in the data repairing section,BPNN is used to map the correlation between FCCU data.The BPNN is trained by learning the loss function between the model output and the actual output,and then the data is patched.The data cleaning work carried out in this thesis aims at the data of FCCU blowdown.It can effectively improve the data quality and provide reliable data support for the later prediction system.