Data-driven Technique And Its Application In Polyproylene Production Process

Process monitoring and fault diagnosis of polymerization process, optimization ofpolymerization process and prediction of polymer important quality index are theinterdisciplinary field with many unsolved and challenging fundamental research topics andpractical applications, relating to chemical reaction engineering and process system engineering.But polymerization process is high complexity, high coupled and high nonlinearity, this lead theresearch for polymerization process very difficult. Because of the rapid improvement ofcomputer techniques, massive data can be obtained by database, so the so-called ‘data-driven’technique how to drawing information from data has been getting more and more attention.Aiming to the shortage of traditional data-driven techniques, some improved data-driventechniques are proposed, such as stacked neural networks, improved multi-scale principalcomponent analysis and improved chaotic particle swarm optimization. And the improveddata-driven techniques are applied in polypropylene production process in this paper, such asprediction of polypropylene melt index, process monitoring and fault diagnosis of propylenepolymerization process and optimal grade transition of polypropylene production process. Theresults obtained in this work are summarized as follows:(1) A modeling approach based on stacked neural networks is proposed. Single neuralnetwork model generalization capability can be significantly improved by using stacked neuralnetworks model. Proper determination of the stacking weights is essential for good stackedneural networks model performance, so two methods about determination of appropriate weightsfor combining individual networks are proposed. Melt index is the most important parameter indetermining the polypropylene grade. Since the lack of proper on-line instruments, itsmeasurement interval and delay are both very long. This makes the quality control quite difficult.Application to real industrial data demonstrates that the polypropylene melt index can besuccessfully estimated using stacked neural networks. The results obtained demonstratesignificant improvements in model accuracy, as a result of using stacked neural networks model,compared to using single neural network model. (2) For enhancing preformance of process monitoring and fault diagnosis, an improvedmulti-scale principal component analysis (MSPCA) is proposed. Considering the non-stationaryand random nature of data in the process industry it contains different noises inevitably. Based onthe characteristics of wavelet analysis, this paper proposed an improved method which combinesmultiple wavelet transform with a new threshold function. The data collected from the industrycondition are processed by means of the improved wavelet threshold denoising method. Usingwavelets, each variable is decomposed into approximations and details at different scales.Contributions from each scale are collected in separate matrices, and a principal componentanalysis model is then constructed to extract correlation at each scale. According to thesimulation of propylene polymerization, and comparing the improved MSPCA with traditionalMSPCA, it shows that the improved MSPCA has enhanced the accuracy of process monitoringand fault diagnosis.(3) Aiming to improve the performance of standard particle swarm optimization algorithm,an improved chaotic particle swarm optimization algorithm is introduced. Chaotic searching isintegrated into particle swarm optimization algorithm. Judgment and handling mechanism oflocal convergence is developed. It greatly enhances the local searching efficiency and globalsearching performance of algorithm. A model of grade transition for the industrial multi-reactorpolypropylene production process is conducted according to the Spheripol technique. The modelof grade transition is solved by using both improved chaotic particle swarm optimizationalgorithm and traditional chaotic particle swarm optimization algorithm. The results show thatthe proposed improved chaotic particle swarm optimization algorithm is superior to thetraditional chaotic particle swarm optimization algorithm one in the optimization efficiency andglobal performance.(4) Finally, the research finding is concluded, and pointed out some future research areas.