Research On Soft-Sensor Method For Arc Current And Voltage Of Ladle Furnace

The arc current and arc voltage are the input of the electrode regulator which is one section of the ladle furnace control system. The output signal is calculated by the electrode regulator based on the value of the arc current and voltage. The accurate measurement for the arc current and voltage has theoretical and practical significance to improve the performance of the electrode regulator, to produce high-quality steel and to increase the production efficiency. But how arc current and arc voltage to be measured accurately and economically is still a problem of present stage.The thesis firstly describes the development and production process of the ladle furnace, the existing measurements for the arc current and voltage are researched. The large arc current is hard to measure directly, and the measurement cost is high. To aim at these shortages, soft-sensor methods are established accessing to the economic and accurate measurement for arc current and voltage.Based on the energy conservation and the electric arc physical knowledge, a new time domain electric arc model and a new power supply system model are developed. Through the model simulation, the effectiveness of the established model is verified. Due to the clear nonlinear correlation of input data, KPCA is used to reduce the complexity of soft-sensor model and cut redundant information efficiently.One soft-sensor model for arc current and voltage based on BP Neural Network is established firstly and the model is trained. The simulation results show the generalization of BP Neural Network is well as a whole. To aim at the defects of the steepest descent in slowly converging and easily immerging in partial minimum frequently of BP Neural Network resulting in the great prediction errors of individual points, the genetic algorithm is brought to optimize the initial weights and thresholds. Then, one soft-sensor model for arc current and voltage based on GA-BP Neural Network is establishedFinally, the analysis and comparison are undertaken between the two models. The result shows that the BP Neural Network model based genetic algorithm is much better than the single BP Neural Network model on convergence rate, model accuracy and generalization ability. The effectiveness of soft-sensor methods for arc current and voltage of ladle furnace is verified through the simulation results.