| In order to improve economic efficiency,product quality and reduce energy consumption and emissions,it is necessary to ensure the safe and stable operation of the process,of which the safety of industrial production processes is particularly important.Process faults and alarm floods are important causes of low production efficiency and even safety accidents.However,the data collected by the actual industrial process is complex,with strong nonlinearity and non-Gaussian characteristics,which poses a huge challenge to the traditional process monitoring based on data classification.At the same time,the existing alarm system is inefficient,and the flooding of alarms can easily lead to industrial accidents.How to suppress the overflow of alarms and reduce the subsequent alarms caused by abnormal propagation is also an urgent problem to be solved.Aiming at the characteristics of complex,strong nonlinear and non-Gaussian nature of actual industrial process data,a new nonlinear parametric dimensionality reduction technique(parametric t-SNE)is proposed to extract the characteristics of industrial fault data and establish a fault classification model,which defines the parameter optimization index is used to determine the optimal parameters of the model,and the k-nearest neighbor classification algorithm(KNN)is used to calculate the extracted features to achieve the optimal fault classification with high accuracy.Compared with traditional fault classification methods such as fisher discriminant analysis(FDA)and local linear Index exponential discriminant analysis(LLEDA),this method retains the nonlinear structure of high-dimensional fault data in the low-dimensional feature space during the dimensionality reduction process,and can better distinguish non-Gaussian nonlinear industrial fault data with only a small number of features,so it has the advantages of high clustering accuracy and low uncertainty.In order to solve the problem of alarm flood caused by subsequent alarms caused by continuous propagation of anomalies,an improved incremental causal pattern mining algorithm is proposed to effectively mine the frequent alarm patterns that cause alarm flood.According to the timing and delay characteristics of alarms caused by abnormal propagation,alarm events and frequent alarm modes with causality are redefined,and time constraints are introduced to improve the prefixSpan algorithm,which greatly improves the effectiveness of causal alarm mode mining.Simultaneously,in order to improve the online application efficiency of this method,an incremental mining strategy is given,which can dynamically update the alarm patterns without rescanning the entire updated database.This method effectively identifies frequent alarm patterns,guides the operator to analyze the alarm priority,suppress the subsequent associated alarms caused by the initial alarm,and solve the problem of alarm flooding to a certain extent.The effectiveness of the two methods proposed in this paper was verified by MNIST data set experiment,synthetic data set,TE process data set and penicillin fermentation process data set. |
