Research On Dynamic Modeling For Gas Mixing Process Based On Subspace Method

In the initial stage of control system, the established models are generally able to reflect the mathematical properties of industrial objects and meet the needs of security and stability in control. However, due to the volatility in actual working conditions and kinds of confounding factors, the problems of traditional off-line modeling methods are facing can been reflected in poor generalization ability and lack of online update mechanism, which would lead to model mismatch and decrease in model accuracy. Therefore, it is important and necessary to build online models for industrial processes, which can realize the adjustment of model parameters when facing different working conditions, and finally guarantee the models to accurately reflect the dynamic characteristics of industrial objects.Gas mixing process is a typical complex industrial process. In this paper, based on industrial data, the model mismatch of gas mixing process is analyzed. By building nominal model, model out in different time is predicted. The results show that due to the changed working conditions, the model fitting degree is decreased and model mismatch appears as time goes on.Aimed at the problem of model mismatch, a dynamic modeling for industrial process based on subspace method is proposed. On the basis of static modeling, a model evaluation mechanism is introduced into dynamic subspace modeling method, including the state matrix mismatch analysis and model predictive output analysis. The former focuses on building a mismatch coefficient, while the latter analyzes the difference of model output and true value. By analyzing the model of mismatch of gas mixing process, the result shows two methods can both reflect the characters of model mismatch. Then, a structure of subspace dynamic modeling method is proposed.The previously proposed subspace dynamic modeling method is used in gas mixing process. First, on the basis of analyzing the practical operation and data, the mechanism models of the gas flow of blast furnace and coke oven are acquired by applying butterfly valve flow formula. Then data-driven subspace identification method is used to set up the4-input and1-output mixing gas pressure model. At last, through analyzing the characteristics of the variables of each sub-model, the two models are integrated. At last, through comparing the difference between predicting results and true values, once the difference exceeds model evaluation index, a dynamic model is rebuilt to track new condition.The simulation experience shows the effectiveness and verification of the dynamic subspace model of gas mixing process, which will provide a precise model for controller design. Research on subspace dynamic modeling method has important implications for those processes with similar characteristics.