The Soft-sensoring Technology Based On Process Neural Network In Chemical Process

Quality control is an important part in the industrial process, but someimportant parameters are difficult to be measured by the hard sensors in alarge number of real production process, which makes it difficult to achieveonline monitoring of the process, and then difficult to achieve online controland optimization operation. A factory of high-density polyethylene (HDPE) isused as an example to research on the soft-sensing and optimizationtechnologies in industrial process.HDPE is a generated high crystallinity, non-polar ethylene copolymerthermoplastic resin, it is produced by the copolymerization of Ethylene. Dueto the excellent physical and chemical properties, it mainly has been applied tomany areas. However, compared with developed countries and regions,polyethylene technology in our country is still a certain gap behind.Polymerization process is the core of the whole polyethylene productionprocess. The application of process modeling, control and optimizationtechniques to guide production operation has become the urgent demands forpolyethylene production enterprises in China.The main research work in this paper is as follows: First, this paper explains the purpose and significance of the research,introduces the soft sensor technology and the main modeling method, whichfocuses on the modeling method based on data-driven technology, andprovides an overview of the current status of research and application of softsensor technology in the world.Second, the output error function is improved and then a soft sensormodeling method based on improved process neural network is proposed.Using a moving window technique to determine the time series containingmost of the information; and then use the improved process neural network toestablish the soft-sensor; at last, adjust the soft sensor to achieve continuoushigh-precision estimates.Third, we start with polymerization mechanism analysis, and establishthe model of melt flow index based on the proposed PNN soft-sensingtechnology, then achieve real-time monitoring of quality indicators. HDPEsimulation examples testify that the proposed method is able to adequatelyprovide on-line estimates of the polymer’s melt index with high predictionaccuracy and good tracking performance, while the adaptive features canensure the reliability of its online application. which is of great value in actualindustrial production control and optimization operation.At last, on the basis of the establishment of soft-sensing models of meltindex, this article uses advanced optimization algorithm to determine theoptimum operating scheme of the polymerization process, and seek the optimal operation point of the process, providing a reference to theoptimization of the industrial control, which play a significant role onguaranteeing products quality, reducing production costs and then improvingthe economic efficiency of enterprises.