Abnormal Detection Of Aluminum Electrolytic Cell Based On Generative Model

The problem of unbalanced time series samples is common in industrial applications.The number of samples under normal conditions is much larger than that under abnormal conditions.Especially in the electrolytic aluminum industry,it is very difficult to find abnormal samples in industrial production because experts do not have a clear criterion to judge abnormalities.In recent years,Generative Adversarial Networks(GANs)have become more and more popular in the field of anomaly detection.In this paper,we use the ability of Generative Adversarial Networks(GANs)to model the distribution of complex high-dimensional images,and propose an self-adaption AAE-GAN method.Finally,based on anomaly detection,clustering algorithms are used to classify anomalies,so that experts have a clear understanding of anomalies,and can make appropriate decisions to reduce losses.The specific research work mainly includes the following:1.Aiming at the problem of the imbalance of time series samples in the field of aluminum electrolysis,using the ability of GAN to learn the distribution of complex high-dimensional images,a new anomaly detection method based on GAN is proposed.In order to increase the training accuracy of the model,this paper proposes a time series anomaly detection method self-adaption AAE-GAN.The model only needs normal samples during the training process,which is a more realistic network than other existing networks,because in actual industrial scenarios,the number of abnormal samples is usually insufficient.The model judges sample category by the anomaly score.The anomaly score is composed of sample reconstructed difference.2.In order to make experts have a clear understanding of anomalies and can make appropriate decisions to reduce losses,this paper proposes a clustering acceleration algorithm based on hierarchical density proposed by Mcinnes^[114]to classify all abnormal samples.This new algorithm is improved on the basis of^*,^*is better than the popular DBSCAN algorithm and the existing density-based clustering algorithm.This makes the accelerated^*algorithm become the default choice based on density clustering.3.Based on the above research,this paper designs and implements the application and analysis system of the aluminum electrolytic cell state abnormality detection model based on the generative model in the aluminum electrolytic industry.The system includes a variety of display methods for aluminum electrolytic cell production data,the display of the abnormal detection results of the aluminum electrolytic cell state and the display of the abnormal classification results of the aluminum electrolytic cell state.