Research On Support Vector Machines Of Soft-Sensor Measurement Model For Aluminum Strip's Grain Fineness Number

In production process of aluminum electromagnetic casting-rolling, aluminum strip's grain fineness number is one of the most important guide lines to scale aluminum plate's quality. However, aluminum strip's grain fineness number could not be detected on-line and real-time, people use metallographic analysis method to realize off-line detection which results in the large delay and influence the production and detection of aluminum plate. Therefore, the study of an quickly and accurately detection method to predict the data of aluminum strip's grain fineness number, can improve the production quality which in aluminum electromagnetic casting-rolling.Based on the analysis of the current production status and methods in which Aluminum electromagnetic casting-rolling, and investigate the electromagnetic factors and casting factors which influence aluminum strip's grain fineness number, in allusion to complexity of production which in aluminum electromagnetic casting -rolling, choose appropriate assistant variables, such as roll diameter, roll-casting speed, cooling water temperature, current amplitude, phase reverse cycle, harmonic component, and study a soft sensor model of aluminum strip's grain fineness number which based on Support Vector Machine. After build relations of assistant variables and dominant variables, we use simulative software to training data group and build model to carry out on-line and real-time prediction of aluminum strip's grain fineness numbers. In allusion to complex process of aluminum electromagnetic casting-rolling system, we discuss kernel function and restrict terms' influences to predict precision and generalization. Whereafter we adoptε-SVR training arithmetic and build a aluminum strip's grain fineness number model which based onε-SVR arithmetic. Because of the restrictions ofε-SVR arithmetic, we adopt v-SVR training arithmetic and build a aluminum strip's grain fineness number model which based on v-SVR arithmetic. After compare with the other model, we draw the conclusion that the last model has better is more feasible for aluminum strip's grain fineness number model.The simulation results show that the aluminum strip's grain fineness number model which based onε-SVR arithmetic is restricted by s parameter and its restrictions, result in large calculated response. Otherwise, the aluminum strip's grain fineness number model whith based on v-SVR arithmetic is more feasible, for it simplified the model parameters and calculation, shortened the response time, administer to detect the aluminum strip's fineness number in the production of aluminum electromagnetic casting-rolling.