Study On Soft Sensor Method For Process In Producting Acetylene By Pyrolysis Of Coal In Plasma

Coal plays an important role in the vigorous development of China's chemical industry.The development of coal chemical industry has effectively alleviated the dependence of China's chemical industry on oil and natural gas.Acetylene is an important basic chemical raw material.In the industry,the traditional methods for preparing acetylene include hydrolysis calcium carbide method and methane oxidation method.However,these methods have high costs,serious pollution,and long processes,and it is difficult to obtain good economic benefits.The plasma cracking coal to acetylene process has the advantages of cleanliness and high efficiency.With the continuous development of this process technology,coal to acetylene has gradually become a trend to replace the traditional process.However,the coal to acetylene process has unknown mechanisms,serious section coupling,time lag,and serious coking problems,which have not been well solved yet,resulting in the current reaction site being unable to run stably for long periods.Realizing the real time measurement of key quality parameters in the reaction process is the first task to solve these problems.At present,most of the research on coal to acetylene is still in the stage of chemical process optimization and mechanism exploration.This paper starts from the perspective of soft measurement modeling and makes real time predictions of two key performance indicators,including acetylene concentration and coke thickness.A data driven soft measurement modeling method is proposed to perform real time prediction of acetylene concentration.Also,the hybrid mechanism and data-driven method are used to predict the coking thickness and achieve high prediction accuracy.Both have achieved high prediction accuracy,laying an important foundation for subsequent research on the whole process control.The main work and innovations of the article are as follows(1)Summarize the research status of plasma to coal acetylene and soft sensor algorithms,and propose the research content of this paper for field measurement problems.The mechanism of acetylene generation in the reaction process is analyzed and discussed in detail,then the auxiliary variables for soft sensor modeling of acetylene concentration are determined,and an improved gradient boosting integrated tree model XGBOOST is proposed to model the acetylene concentration.This method has strong nonlinear expression ability.It relies on rules and can realize automatic feature selection,which can effectively reduce the prediction bias of the model.Finally,experiments show that the method has a good prediction effect.(2)Aiming at the fact that traditional modeling strategies fail to make full use of the large amount of unlabeled data in the field and do not consider the dynamic characteristics of the process,a semi-supervised learning based on improved LSTM model is proposed,which can effectively solve this problem by means of sequence modeling.At the same time,the convolution unit and multi-layer structure are introduced to extract the original information in depth,and the problem of poor long-term modeling effect of LSTM is solved by the Attention mechanism.Based on real data,it is verified that the proposed method can meet the needs of industrial site prediction.(3)After a detailed analysis and discussion of the coking mechanism,combined with the characteristics of the field reaction device,a coking thickness measurement method based on the mechanism model was proposed.Mechanism modeling relies on hydrodynamic models.Considering that the mechanism model is ideal,and other factors are not considered,a data driven small sample modeling model SVR is introduced to perform a secondary correction on the prediction bias of the mechanism model,which can obtain a soft sensor model with a mixed strategy.Simulation experiments show the effectiveness of the hybrid modeling strategy.