Research On Prediction Analysis And Application Of Time Series Data For Production Proces

With the development of information technology,data grows in massive form,and the cross-integration of multi-field data with the help of data mining technology is conducive to more accurate,more efficient,and more scientific management and decision-making.At present,data predictive analysis and practical application is a huge challenge for data mining.It is of great significance to deepen and develop the intelligent manufacturing system to establish a time series prediction analysis model oriented to the production process,and to develop an intelligent manufacturing system for time series data analysis,trend prediction,independent analysis and decision-making oriented to the technological process.In order to improve industrial production efficiency and monitor the status of production indicators in real time,predictive analysis of production process time series data has become the focus of research.However,traditional time series forecasting methods are prone to defects such as poor robustness and low precision,and at the same time,there is a lack of efficient and accurate time series forecasting methods in the analysis of production process data.Therefore,for the production process time series data prediction and practical application,the main research contents of this paper are as follows:(1)In terms of the characteristics and models of production process time series data,this paper first analyzes the relevant characteristics of production process time series data,summarizes the classic time series data representation models,and studies the intelligent optimization algorithms commonly used in time series data forecasting.And neural network prediction model,laying a theoretical foundation for the subsequent production process time series prediction.(2)Aiming at the short-term prediction of univariate time series data of production process,this paper proposes two short-term prediction models based on Adaptive Fuzzy Neural Inference System(ANFIS),namely ADE-ANFIS accurate prediction model and GNG-ANFIS efficient prediction model.In the ADE-ANFIS model,an improved adaptive differential evolution algorithm is used to optimize ANFIS,which improves the prediction accuracy of the model;in the GNG-ANFIS model,GNG is used to dynamically track the original data,filter the data,and reduce the training time of the model..Combined with the time series data of impurity iron content in the aluminum electrolysis process,the performance verification of the two models is completed.(3)Aiming at the problem of long-term trend prediction of multivariate time series data of production process,this paper proposes a prediction model based on deep learning.Since the complex relationship between multivariate time series will affect the prediction effect,the wavelet correlation analysis method is used to analyze the multivariate time series and select the most relevant variables.At the same time,in order to make up for the defect that the shallow model cannot accurately mine potential information,deep LSTMs are built.,which improves the prediction accuracy.Finally,combined with the time series data of production factors such as bath temperature and voltage in the aluminum electrolysis production process,a long-term forecast of the trend of aluminum output is carried out.(4)Based on the research of GUI aluminum electrolysis process time series data prediction system,with the help of the above research results,build an object-oriented aluminum electrolysis time series data prediction analysis and application system,including:the production principle of aluminum electrolysis,production collection time series data Analysis,short-term forecast of impurity elements in electrolytic cells in production,and long-term trend forecast of aluminum output,and finally feedback the results to managers/decision-makers in a timely manner.