Research On The Quality Prediction Of Chemical Products Based On XGBoost And LSTM Models

With the accelerating pace of digitalization,networking and intelligence in global manufacturing,China's industry is facing the problem of transformation and upgrading.The deep integration of information technology and traditional industry is the main development direction of the fourth industrial revolution.The contemporary industrial manufacturing process is becoming more and more complex,mainly showing the following characteristics: the scale is enlarged,the processing steps are complicated,and the industrial manufacturing process forms a complicated system;the abnormalities occurring in the manufacturing process often cause huge losses.Unfortunately,even high-end equipment in industrial manufacturing processes cannot avoid the operational anomalies that arise during operation,and the large number of human interaction processes that exist during the manufacturing process can cause human error to trigger anomalies throughout the manufacturing process.It can be seen that if the quality of the product or the abnormality in the manufacturing process can be predicted and prevented in time,the efficiency of the entire production process will be improved,thereby ensuring the quality and service of the product.At present,the popular quality control method used by most industrial enterprises is to invest a large amount of maintenance costs to ensure the normal operation of the manufacturing process.However,some industrial enterprises have begun to explore how to use statistical machine learning and artificial intelligence to learn patterns from a large amount of data accumulated in the production line,so that they can predict the quality of industrial manufacturing processes in advance,and discover problems in the industrial manufacturing process in time,and ensure that the industrial manufacturing process can be carried out quickly and efficiently.This paper focuses on the quality prediction of chemical products,and specifically solves the problem of time series prediction of industrial big data.The author used two data sets with the same source(both from the real chemical enterprise production environment)but different processing methods to explore the above problems from two completely different research perspectives.The first angle is for the data set processed by the revolving door compression algorithm.This paper builds the model based on feature engineering and integrated learning which called extreme gradient boosting.The other angle is for the original long time sequence(here the long time sequence refers to each chemical batch has a large number of recording points at the moment).This paper first uses an encoding-decoding technique similar to machine translation to convert it into a short sequence,and then does not rely on any additional manual processing,using long and short-term memory networks.The deep learning model models it.Finally,the author integrates the above two angles of the modeling framework to generate the final prediction model.The experimental results show that the use of the integrated learning framework to timely control the production process to achieve quality control results outperforms the traditional quality control methods of enterprises in terms of greatly reducing the operating costs of industrial enterprises,and significantly improving the production efficiency of enterprises.The innovation of this paper is to combine the popular machine learning method with the latest deep learning technology,and comprehensively utilize the advantages of both to achieve more accurate and intelligent chemical product quality prediction.