The Study Of Chemical Engineering Modeling And Predictive Method Based On Neural Network

Chemical process modeling is a main field of chemical engineering process and plays a fundamental role in process simulation, optimization and control. However, because most chemical processes have very complex mechanisms, strong-coupling and high severe nonlinearities, it's hard to establish mechanism model, which is the reason why empirical modeling is adopted. Artificial neural network models do not need to consider the mechanism of chemical processes and have an excellent effect on nonlinear problems. It's a typical kind of empirical model and has been applied to many chemical problems. In this paper, based on the actual measurement of experiment data, three methods are separately used to set up dry gas-to-ethylbenzene reactor model, such as BP neural network, GA-BP neural network and support vector machines. The main content of the dissertation is as follows:1,A brief introduction is made about BP neural network. It mainly contains the basic principle and the algorithm description about BP neural network. A process that predicts the output temperature in dry gas-to-ethylbenzene reactor by BP neural network is stated in detail. The selection of input and output data and the method for determining the structure in neural network are discussed mainly. Then, establish a predicting modeling and predict about training set and testing set respectively. As a result, the relative coefficient about the predictive result is above 80%, which cites that the modeling accuracy has reached the requirement. Through the application of BP neural network in dry gas-to-ethylbenzene reactor, the advantages of BP neural network has been concluded.2,A brief introduction is made about GA, and the ideas about optimizing of BP neural network by GA. Applying GA-BP theory into predicting the output temperature in dry gas-to-ethylbenzene reactor, a modeling structure of 10-18-1 is selected with a population size of 10, a evolution of 50, a crossover probability of 0.4, and a mutation probability of 0.2, and then establish the output temperature predicting modeling in dry gas-to-ethylbenzene reactor. The detection of modeling on the measuring data is conducted, and the relative coefficient of measurement can reach 83.6103%, which illustrates the available of modeling. Via the comparison of predictive results between BP and GA-BP, the modeling stability and modeling accuracy of GA-BP both took a obviously positive position. 3,A brief introduction is made about support vector machines. Applying support vector machines theory into predicting the output temperature in dry gas-to-ethylbenzene reactor, selectingε-SVR as the type of SVM and RBF as kernel function, choosing the best parameters C=5.6569,γ=0.015625 by cross-validation method and then establish the output temperature predicting modeling in dry gas-to-ethylbenzene reactor. The detection of modeling on the measuring data is conducted, and the relative coefficient of measurement is above 90%, which illustrates the available of modeling. At last, we made a meaningful discussion about the prediction error. Via the comparison of predictive results between SVM and GA-BP, the modeling stability and modeling accuracy of SVM both took a obviously positive position. SVM was a technique even more suitable for practical application.In the end of this dissertation, we make a summary and describe the further works.