中圖分類號(hào)：TP995
文獻(xiàn)標(biāo)識(shí)碼：A
DOI：10.19358/j.issn.2097-1788.2023.07.014
引用格式：馬勇，楊敏，朱琳.面向多維數(shù)據(jù)的異常點(diǎn)檢測(cè)模型設(shè)計(jì)［J］.網(wǎng)絡(luò)安全與數(shù)據(jù)治理，2023，42（7）：85-90.

Design of outlier detection model for multidimensional data

Abstract： In order to quickly and accurately mine outliers in the big data environment and ensure network security, we propose a design scheme for multidimensional data oriented outlier detection model. In this scheme, the long short memory network (LSTM) is used to store multidimensional data in any period of time, and the graph convolution network is used to extract the complete data structure. At the same time, penalty parameters and mean square error are added to narrow the range of outliers. In addition, we also use the encoder and decoder to build a variational self encoder function model, so that it can interpret the normal data sub features, and calculate the data anomaly measurement through the coding reconstruction loss function, so as to achieve outlier detection. After experimental verification, this method exhibits high detection accuracy and operational efficiency, and has high application value.

Key words : coding loss function; variational self encoder; abnormal point detection; long and short term memory network; multidimensional data