Research On Soft Sensor Methods For The Current And Voltage Of Electric Arc Furnace

Electric arc furnace steelmaking has nearly a hundred year of history,due to its ability to make use of scrap resources,small energy consumption,short steelmaking process,the development in recent decades is particularly rapid,has been more and more widely used.The most important part of the EAF control system is the electrode controller.The input of the electrode controller is the arc current and the arc voltage.Therefore,accurate and rapid measurement of the arc current and arc voltage is of great significance to the stable and efficient operation of the EAF.However,due to the arc current up to tens of thousands of amperes,the arc voltage can not be directly ground measurement and other reasons,how to measure them economically and accurately is still the current problems.Soft sensor technology provides a new idea for measuring the parameters that are difficult or even impossible to measure in industrial hardware meter and has become one of the new hot spots in the field of process control.In this thesis,a 70t electric arc furnace in an iron and steel company in Xuzhou,Jiangsu Province is taken as the research object.The soft sensor methods of arc current and arc voltage are studied,and the soft sensor models of arc current and arc voltage are established.The experimental results show the effectiveness of the method.In this thesis,the equipment and process of EAF are briefly described.The relationship between arc current and arc voltage and the primary current,voltage and total active power of three-phase transformer is given based on the principle of transformer.Auxiliary variables in soft-sensor modeling were identified which laid the foundation for the establishment of the data model;For partial Least Squares global model,there are some shortcomings in dealing with data nonlinearity and data mutation.Based on Just in Time Learning,an electric arc furnace currentvoltage soft-sensing model is built,and the actual production data is used to simulate the results show that the model has high predictive accuracy and good tracking performance.However,this method of on-line modeling consumes too much time and timeliness when calculating the parameters,and it is difficult to meet the requirements of the rapidity of the EAF smelting process;Aiming at the shortcomings of the above methods,an integrated learning strategy(Ensemble Learning)is used to establish the current-voltage soft-sensing model of EAF.Using the Moving Window strategy to complete the partition of the local area,the establishment of local models and predict the output,based on Bayesian integrated learning method to local adaptive output integration.The simulation results of the model show that the model has better prediction accuracy than the model established by Just in Time Learning strategy,and the prediction time-consuming is much less,which is in line with the standard of engineering application.The method has certain reliability.