Research On Equipment Portrait Based On Time Series

As an important production equipment in the aluminum industry,the electrolytic cell should make full use of massive production data to promote industrial digital transformation while studying how to increase production capacity.This is the focus and difficulty that the electrolytic aluminum industry needs to solve.Equipment portrait is a technology that uses various data mining methods to extract practical information from various data generated in the equipment production process as a portrait label,so that the quality of equipment production can be grasped more objectively and accurately.It lays the foundation for the realization of quality and safety monitoring,trend forecasting and model optimization.At present,there are few researches on equipment portrait technology based on time series analysis,and there is a lack of systematic mining methods to present data value;moreover,the process of aluminum electrolysis is relatively complicated,and there are interactive problems in the actual production process.In order to extract features from highly correlated and high-dimensional data and expand multi-dimensional time series data horizontally,this paper proposes a method combining statistical models and neural network models to predict multivariate time series while extracting feature labels;longitudinally analyze the data of multiple electrolytic cells Time series correlation,the feature labels are extracted from the data set according to the two rules of abnormal points and trends,and model clustering and improved partition clustering are used to classify data points into specific categories according to labels to generate time series based Electrolyzer industrial equipment portrait.The main research contents are as follows:In the time series based equipment portrait label selection stage,in order to obtain the nonlinear and hidden features in the aluminum electrolysis time series,the vector autoregressive moving average model is used to fit the multivariate parameter features;combined with the Time2 Vec vector embedding time form as the enhanced data of the neural network source to automate the creation of feature engineering and generalize deep learning techniques.In order to solve the problem of catastrophic forgetting,Seq2 Seq is used to adjust the structure of neural network to predict the parameters of electrolytic aluminum.The experimental results show that the effect of the model is significantly improved after adjusting the network structure,which proves its practicability in the field of engineering applications.For visual analysis and exploration of time series data,the Pa CMAP dimensionality reduction technique is used to reduce the number of features to two dimensions.In the extraction of anomaly detection labels,separate the mixed Gaussian distribution probability in aluminum electrolysis data to improve the representation of data to increase the accuracy of clustering results and improve the effect of anomaly detection;In the extraction of trend labels,a method combining dynamic time warping and parameter correlation weighting is proposed to collect time series trends of similar shapes to improve equipment portraits.The experimental results show that the label information extracted from the multi-dimensional data can reflect the status of each stage of the equipment in detail,and the integration of the portrait technology into the industrial control system can effectively improve the management and regulation of the equipment.