Research On Soft Sensor Of Industry Process With Multiple Working Conditions Based On Transfer Learning

In modern industrial processes,due to constraints such as production environment,testing technology,and equipment costs,it is difficult to achieve online real-time detection of some key quality variables,which are often crucial to process control and product quality.Soft sensor technology can indirectly achieve effective estimation of dominant variables by establishing a mathematical model between easily measurable auxiliary variables and dominant variables.With its economic and reliable online testing,rapid response,and easy to achieve real-time monitoring and control of product quality,it has become a hot research issue in the field of industrial process control and testing.However,due to the multiple working conditions of actual industrial production processes,the performance of traditional soft sensor models decreases as the working conditions change.Therefore,this paper considers the problem of performance degradation of soft sensor models due to changes in data distribution under multiple operating conditions,and conducts research on soft sensor modeling methods based on transfer learning.The main research contents of this paper are as follows:(1)Aiming at the problem of poor prediction accuracy in establishing a single global model for traditional soft sensor modeling methods under multiple working conditions,and requiring that training data and test data must obey the same distribution,an integrated soft sensor method based on transfer component analysis is proposed.Using transfer component analysis to solve the shared feature mapping matrix between samples,adapting the edge probability distribution of training data and test data;Based on the Gaussian mixture model,the adaptive training data is clustered and divided,and combined with the partial least squares algorithm to establish an integrated model of sub models to complete the prediction of dominant variables.The application simulation results of numerical simulation and penicillin platform data show that the proposed method can effectively improve the prediction accuracy and generalization ability of the soft sensor model under multiple working conditions.(2)Aiming at the problem that traditional transfer methods tend to lose data information and reduce the accuracy of soft sensor models during overall transfer,a soft sensor modeling method based on a local transfer modeling framework is proposed.Combining the idea of geodesic flow kernel and local modeling,firstly,K-means clustering is used to extract local feature information,and the training set is divided into several sub source domains.Correlation analysis and discrimination are performed by measuring the Euclidean distance between the sample to be tested and the cluster center of each sub source domain,so that transfer learning is performed in the sub source domain with the highest correlation to the sample to be tested.Then,based on the geodesic flow check,the sub source domain data and the sample data to be measured are feature mapped,and a prediction model is established using a partial least squares algorithm.The simulation results of numerical examples and industrial data show that the proposed method can fully utilize the local information of process data and effectively improve the accuracy of the soft sensor model under multiple working conditions.(3)In order to better apply the subject research,a data analysis and soft sensing system was designed and developed based on the main research work.The system takes soft sensor modeling and prediction as its main function,and aims to meet the concise and efficient use needs of industrial sites.It uses front-end technology to design a web page as a visual operation part of the system,and accesses the system through a web address,reducing the threshold for deployment and use;Utilize the high computing power of cloud servers to achieve system data analysis,soft sensor modeling algorithm invocation,and efficient communication with databases,ensuring stable system operation.At the same time,it provides functions such as modeling algorithm library,historical data display,and system permission management,further improving the system usage experience and scalability.