A Soft Sensor Method For Batch Processes Based On State Estimation

Batch processes is an important industrial production method,which is widely used in fine chemicals,biopharmaceuticals,microelectronics and other fields.The key parameters in the batch processes are closely related to the quality of the product,so it is necessary to obtain the online measurement data of the key parameters in the batch processes,but limited by sensing technology,key parameters such as biomass concentration and substrate concentration are not easy to be measured online.By establishing an batch processes soft sensor model,the key parameters that are difficult to measure can be estimated online,and the state estimation method can be used to solve the soft sensor model,which can reduce noise interference and improve estimation accuracy.However,the batch processes has the characteristics of multiple stages,and the overall soft sensor model cannot fully describe the dynamic characteristics of each stage of the batch processes.The lower accuracy of the overall soft sensor model affects the soft sensing of key parameters of the batch processes.Existing methods establish a segmented soft sensor model for batch processes,and the establishment of a segmented soft sensor model for an batch processes needs to consider the problem of stage division and transition stage modeling,which makes the implementation of segmented soft sensor modeling for an batch processes complicated.Therefore,in the case of using the overall soft sensor model of the batch processes,it has important application value to improve the estimation accuracy of the soft sensor method for the batch processes based on state estimation.In this dissertation,a key process event node matching strategy is constructed,and a state estimation-based soft sensor method for batch processes is studied.First,in order to solve the problem that when the dynamic time warping method is used for batch data online warping,the amount of measured value vector sequence data continues to increase with the progress of the reaction,resulting in a large amount of online warping calculation and affecting real-time performance,a Variable Window Dynamic Time Warping(VWDTW)method is proposed,which introduces window processing and window selection strategy in dynamic time warping,which improves real-time performance while ensuring online warping accuracy.Then,the batch processes measurement equation is supplemented by the relevance vector machine(RVM),and the soft sensor hybrid modeling of the batch processes is realized by combining the mechanism model.A key process event node matching strategy is constructed.The initial value of the state estimate is substituted into the state equation to recursively obtain the state prediction sequence of the batch processes,and the state prediction sequence is substituted into the supplementary batch processes measurement equation to obtain the measurable parameter prediction sequence.Through the online warping of the measured value vector sequence and the measurable parameter prediction sequence,the matching of the key process event nodes predicted by the state equation and the key process event nodes of the actual production process is realized,thereby improving the estimation accuracy of the soft sensor model.On the basis of the established batch processes soft sensor hybrid model and the constructed key process event node matching strategy,the batch processes soft sensor based on state estimation is realized.The experimental research shows that the constructed AWDTW method can improve the real-time performance while ensuring the accuracy of online warping.The constructed key process event node matching strategy can improve the estimation accuracy of batch processes soft sensor based on state estimation.