Research On Anomaly Detection Method Of Process Data Based On Supervised Learning

In recent years,the Internet,machine learning,artificial intelligence,edge computing and other high-end technologies are developing more and more rapidly.In the continuous integration and innovation of technologies,it also promotes the development of China's high-end manufacturing industry.The complex process process means that there may be abnormal operating conditions at every step of the factory operation.The complexity of the industrial site has brought great challenges to the high-quality production of process products.In the production process,every parameter setting may lead to the failure of the product,or even cause the industrial production disaster.Safety in industrial production has aroused more and more people's concern.The industrial control system is the "brain" of the entire industrial production,and the safety of the industrial control system is the focus of the state and enterprises.Firstly,the industrial control system is briefly analyzed in this paper.This paper introduces the application of machine learning technology in anomaly detection of industrial control system.This paper mainly introduces the principle and application of ensemble learning framework in machine learning algorithm.Secondly,the characteristic engineering method of process data is described.The feature engineering processing of data mainly includes noise reduction,data missing value processing and data dimension reduction.The original process data collected in the industrial site contains a lot of noise.Due to the complexity of the process flow and the timeliness of the data collected,the process data collected will have missing data dimensions,and the data imcompleteness will affect the training results of the model.In the third chapter presents several characteristics of data dimension reduction method,this paper introduces the principal component analysis(pca)and stacked the coding method,this paper selects the principal component analysis(pca)combined with integrated rule tree feature selection methods,importance to the characteristics of the data sorting,the importance to select the characteristic of data,to be able to maintain data integrity to the greatest extent.Finally,this paper studies an anomaly detection method of process data based on integrated framework.Collect the data from the industrial site,carry on the characteristic engineering processing to the collected data,adopt the principal component analysis method(PCA)and the integrated rule tree method,carry on the feature selection to the process data,can restore the original characteristic information of the data to the greatest extent.The random forest(RF),Adaboost,XGBoost and SVM were selected as the basic learning machine,and the logical regression(LR)was used as the secondary learning machine to build the anomaly detection model based on the Stacking integrated method.The TE data were used to train the basic learning model and the integrated model respectively,and the experimental results of the integrated model and the experimental results of the basic learning model were compared and analyzed.Finally,the conclusion was drawn that the accuracy rate of abnormal detection of process data was effectively improved,and the false positive rate of abnormal detection of process data was effectively reduced.