Research On Quality Prediction And Operation Optimization Method For Batch Process Based On Process Transfer Model

Due to the advantages of producting a small amount of high-value-added production,batch process have occupied an important position in modern industrial manufacture.In recent years,with the rapid development of data storage and sensor technology,the data-driven method has become an important research direction in the field of quality prediction and operation optimization for batch process.However,building a data-driven model usually requires a large amount of historical data.For some actual production processes,such as a new batch process that is first put into operation,the available modeling data is often insufficient,which may lead to the difficulty in implementing the traditional data-driven prediction and optimization methods effectively.In this paper,the problems in quality prediction and operation optimization of batch processes with insufficient data are studied.The main research contents of this paper are as follows:(1)A JYKPLS process transfer model is proposed to solve the problem of insufficient data and non-linearity of the new batch process.The advantage of this model is that it can make full use of useful data in similar processes to compensate for the lack of modeling data of the new batch process,and then save production cost and improve modeling efficiency.Based on this process transfer model,a new quality prediction method for batch process based on process transfer model is proposed.Firstly,this method uses JYKPLS method to combine similar process data to establish quality prediction model.In online prediction,data estimation approach based on PCA mapping is used to supplement the unknown data.At the end of the production batch,the accumulated actual operation data are updated to the modeling dataset,and the similar process data are gradually eliminated according to their similarities in order to further improve the prediction accuracy of process transfer model.Finally,this method is applied to predict penicillin concentration.The simulation results shown that the method can achieve good prediction results with only a small amount of new process modeling data.(2)Based on the above quality prediction methods,an operation optimization method based on process transfer model is proposed for batch process with insufficient data.Firstly,the optimization method based on process transfer model is used to solve the setting value of operation variables,which can make full use of similar process data to help the new process realize batch-to-batch operation optimization quickly.However,there is often a certain deviation between the process transfer model and the actual process,which leads to NCO mismatch in the optimization model,and ultimately makes the optimization results based on the process transfer model only a sub-optimal solution.To solve this problem,an optimal compensation method based on just-in-time learning and trust region method is proposed in this paper.In this method,the sub-optimal setting value is used as the query point,and the local model based on just in time learning is used to describe the correlation between the compensation value and the increment of economic indicators.Based on this local model,the trust region method is used to solve the optimal compensation value of the sub-optimal setting value.Next,using compensation value to modify the sub-optimal setting value can make the optimization result closer to the optimal.Finally,the proposed method is applied to the operation optimization of cobalt oxalate synthesis process,and the effectiveness of the proposed method is verified.At the end of this paper,the main works of this paper are summarized and the future research contents are prospected.The thesis includes 21 figures,5 tables and 88 references.