Research On Dynamic Soft Sensor Modeling Of PTA Average Particle Size

In order to reduce production cost,monitor process status,improve production efficiency and optimize product quality,real-time measurement and prediction of key quality variables in industrial process is very important.For the complex industrial production environment and the problem that hardware sensors cannot or cannot measure key quality variables,soft measurement method indirectly estimates and predicts key quality variables by constructing an estimation model that takes process variables as input and quality variables as output.Because of its advantages such as easy development,flexible configuration and fast response speed,it has been rapidly developed and effectively practiced in academia and industry.The main research contents of this paper are as follows:(1)Aiming at the problem of low prediction accuracy of soft sensor model due to delay in industrial process,a dynamic soft sensor method based on differential evolution algorithm is proposed.By combining differential evolution algorithm and partial least square method,the delay parameter estimation problem is transformed into a multidimensional linear optimization problem,and the global search ability of differential evolution algorithm is used to solve the optimization problem.The method is applied to PTA(Terephthalic Acid)average grain size soft measurement in PX oxidation process.The simulation results show that the time delay estimation can effectively deal with the delay problem in industrial process and improve the prediction precision of soft measurement model.(2)In order to solve the problem that the learning of soft sensor model is not perfect and the prediction accuracy is affected by using small sample data set when establishing soft sensor model based on artificial neural network,a data enhancement method based on Laplacian noise is proposed.This method expands the limited sample data set,forms more reliable training samples,and alleviates the problem of data set shortage in the process of soft sensor modeling.After the data expansion of PTA average particle size soft measurement training sample,the convolutional neural network is used to build a small sample high-dimension dynamic soft measurement model,and the validity of the data enhancement method is verified.(3)For less data collected in the process of industrial training problems,put forward dynamic soft sensor modeling method based on generating against network,alleviate the shortage of shortage of data set from the modeling method,generating against network by back propagation criterion of updating parameters,the training of the network data don’t need a lot of training and learning,through the rivalry between the generator and discriminant training,Perfect the network model.The generative adversarial network is used in PTA average particle size soft measurement.The effectiveness of the proposed method is verified by comparing with the dynamic soft measurement model based on BP neural network and the dynamic soft measurement model based on small sample and high dimension convolutional neural network based on enhanced data.