Primary tower sampling data gross error discrimination method based on local feature abnormal factors

Abstract

The invention discloses a primary tower sampling data gross error discrimination method based on a local feature abnormal factor, which quantifies the abnormal degree of each piece of sample data by designing a local feature anomaly factor, so that the sample data with larger local feature abnormal factor can be discriminated as gross error. According to the method, multiple pieces of neighbor sample data need to be found for each piece of sample data, then the neighbor sample data serve as a reference data set, local features capable of distinguishing differences between the sample data and neighbors of the sample data are obtained through optimization, and local feature abnormal factors are obtained through calculation. The method does not involve calculation of the Mahalanobis distance, but seeks neighbor sample data of each sample data, and maximizes the difference between the neighbor sample data and the neighbor sample data through the transformation vector, thereby judging whether each sample is a gross error or not through the distance between the local feature and the original point. Therefore, according to the method, whether the sample data is gross error data or not and the number of gross error data in the sample data set can be judged at the same time.

Description

technical field [0001] The invention relates to a method for discriminating gross errors of sampling data, in particular to a method for discriminating gross errors of sampling data of an initial distillation tower based on local characteristic abnormal factors. Background technique [0002] The initial distillation column is the leader of the atmospheric and vacuum unit in the refinery, and its control directly affects the improvement of crude oil yield and the operation stability of subsequent processes. The dry point of the initial top naphtha is an important control index of the initial distillation tower, but there is no suitable online analyzer to measure this index, so it can only be analyzed offline and takes a long time. Therefore, the establishment of the naphtha dry point soft-sensing model can provide guidance for timely adjustment of production and operating conditions. The accuracy of the data of various factors that affect the dry point of initial top naphtha...

AI Technical Summary

Problems solved by technology

Therefore, if the linear correlation between the sampled data is strong, the corresponding covariance matrix will be irreversible

Moreover, the dynamic nature of the sampling data operation of the initial distillation tower will lead to a complex distribution of sampling data, which cannot be generalized simply by relying on the ellipse defined by the Mahalanobis distance

Images