Prediction For Converter Gas Holder Levels Based On Multi-output T-S Fuzzy Model

The converter gas system has the characteristics of multiple-variable, strong coupling and dynamic nonlinearity, which is a system that involves many production and consumption users, delay and limited capacity of the buffer device. The prediction of converter gas holder level is one of the prerequisites for the improvement of converter gas utilization efficiency in iron and steel enterprises. By achieving dynamic balance of byproduct gas system, the production cost is reduced thus leads to enterprise energy saving and consumption reduction. In the converter gas system, the control of gas holder is the base of energy balance scheduling for the workers. If its variation can be effectively predicted by establishing a reliable prediction model of converter gas system, it will provide scientific guidance and decision-making basis for the efficient use of converter gas to achieve dynamic balance of the gas system.Given that a typical converter gas system in iron and steel enterprise has two cabinets, and considering the complex influence factors of the system, this paper proposes a multi-output T-S fuzzy modeling method based on Kalman filter estimation for predicting the dual tank level of converter gas. Considering the converter gas system demand of modern industry that the gas is firstly converted into the two holders and then it is used by the users, this study established a fuzzy prediction model with multiple outputs. The extraction of fuzzy rules is divided into the following two steps. Firstly, the antecedent parameters are identified by using the fuzzy c-means clustering algorithm based on the sample data of historical holders flow value and two holders flow difference data clustering. Then the parameters of membership function are obtained according to the results of clustering center and partition matrix. Secondly, transform the premise parameter identification into overdetermined linear equations. Among them, calculating coefficient of pseudo inverse matrix may produce abnormal solution in the least squares identification parameters. Instead, using the method of constructing the linear equations, the optimal solution of the linear equations is estimated by Kalman filter algorithm. Thus the optimization of the premise parameters is accomplished.In order to verify the proposed model, this paper tests three different kinds of datasets, including Mackey glass dataset with Gaussian white noise, the domestic iron and steel company energy center of converter gas system data about zone Ⅰ and Ⅱ. In order to evaluate the performance of the proposed method, this study compares three algorithms, including the least square support vector machine, T-S model based on least squares parameter optimization method and the T-S fuzzy modeling method based on Kalman filter parameter optimization.. The experiment results shows that the proposed T-S fuzzy model has the advantages of high accuracy and short time consumption over other holder level prediction models. In addition, the proposed method is more suitable for the needs of modern iron and steel enterprises, due to the characteristics of two holders operation of the converter gas system.