Real-time anomaly detection method based on generative adversarial network

Abstract

The invention relates to a real-time anomaly detection method based on a generative adversarial network, and the method comprises the steps: offline training and real-time detection; the offline training comprises the following steps: inputting batch historical normal data acquired from production equipment into a generative adversarial network model after data cleaning, the adversarial network model generating abnormal data and performing identification detection, and deploying the adversarial network model to a real-time detection environment after training is completed; the real-time detection comprises the following steps: collecting real-time data from production equipment, synchronously storing the data into a historical database, and inputting the real-time data into a generative adversarial network model for anomaly detection after data cleaning; when the detection result is normal, marking the real-time data as normal data, and inputting the normal data into the adversarial network model in a backflow manner for incremental training; and when the detection result is abnormal, triggering an abnormal alarm, waiting for manual processing, marking the real-time data as abnormal data when the detection result is confirmed as an abnormal result, and returning the abnormal data to the identification network part of the model for incremental training.

Description

technical field [0001] The invention relates to data supervision and detection technology, in particular to a real-time anomaly detection method based on generation confrontation network. Background technique [0002] The machine learning method of predicting and judging the abnormality of product quality based on the time-series process data collected by the production line has been widely used in manufacturing enterprises. [0003] These methods can be classified into three types by data labeling methods: [0004] The supervised anomaly detection method first marks the production line data as normal and abnormal, and then uses the marked production line data to input neural network or random forest and other supervised machine learning models for training, and then uses the model to analyze the newly collected Production line data for forecasting. In the Chinese patent "abnormal index detection method, device, computer equipment and storage medium", the normal and abnorm...

AI Technical Summary

Problems solved by technology

[0007] The supervised anomaly detection method requires more labeled data to train the model. Since abnormal data is usually scarce in manufacturing production lines, and the labeled data requires professional knowledge, the labeling cost is high. In the actual quality anomaly detection of many production lines When applied, there will be very little labeled data, or even no abnormal data. It is difficult for supervised methods to find enough data to train the model and cannot effectively identify abnormal data.

[0008] In the unsupervised anomaly detection method, it is usually difficult to determine the prior parameters of a specific scene, and it is easy to cause algorithm deviation. Many unsupervised methods need to manually determine the number of categories or the distance threshold between categories. In the case of only normal data, the unsupervised method difficult to function

[0009] Semi-supervised anomaly detection method, when abnormal data is generated from normal data, the training data of the production line usually cannot cover the complete data distribution of normal data

After training with historical data, the production line data often has new changes. The semi-supervised model is limited by historical data, and it is easy to misdetect normal data as abnormal data.

Images