Study On The Prediction Method Of Rubber Particle Concentration Based On Mixed Processing Of Multivariate Information

The rubber industry is an important basic traditional industry in our country’s national economy,and China is the largest producer of synthetic rubber.With the rapid development of automotive,medical and other fields,the demand for butyl rubber and halogenated butyl rubber is increasing,however,it also faces problems and challenges such as few high-end product categories,insufficient output,mainly relying on imports,backward technology and equipment technology,and small profit margins.This paper takes the continuous solution method to prepare bromobutyl rubber production equipment as the research object.Based on the fusion of multi-dimensional information such as production process parameters,equipment operating condition data,product quality data,etc.,for the research on the "soft measurement" method of rubber particle solution concentration.By collecting valid samples in the production process,using feature engineering to extract key features,and establishing a concentration prediction model based on BP neural network algorithm,real-time prediction of colloidal particle concentration is realized,so as to guide the operator to add the slurry ratio before the polymerization reaction.The main work and innovative research results of this paper are as follows:1.The key feature extraction method for rubber particle solution concentration prediction is studied.Combined with the quality characteristics of industrial data,the method of missing value processing,bin division,normalization,and feature combination is used to carry out correlation analysis on production data,and refine the 6 key features: conveyor belt discharge flow,extruder current,extruder motor temperature,extruder grinding head pressure,mixing tank feed pressure,and mixing tank inner wall temperature.2.Propose a scenario-based industrial big data application modeling method.Collect effective samples from massive data on rubber production equipment,processes,operations,equipment operation,quality testing,etc.,and integrate key equipment operating mechanism and manual operation experience,establish a rubber particle concentration prediction model,and carry out model training,experiments,and error analysis.and on-line evaluation to realize real-time "soft measurement" of colloidal particle concentration.3.Based on the above key feature extraction and industrial big data application modeling methods,a 50,000 t/a butyl rubber plant in Shandong Jingbo Petrochemical Co.,Ltd.was selected for application verification.The comparison error between the actual operation results of the rubber particle concentration prediction model based on the BP neural network algorithm and the laboratory calculation results is less than50kg/h,and the relative error can be controlled within 1%,the CPK value of the rubber particle solution concentration has increased from the original 0.75 to 0.86,and the stability of quality is greatly improved.