| China is the world’s largest steel producer and consumer,but also the most concentrated areas of steel investment.As the most important method of ironmaking in the world at present,the improvement of its technology and the study of modeling and control of iron making process have been paid much attention by domestic and foreign experts and scholars.Blast furnace iron smelting is a complex physical,chemical reaction process where liquid iron is transferred from iron ore and other solid state iron compounds through high temperature and high pressure.But due to the complex physical and chemical reaction inside the blast furnace,the coupling of parameters,nonlinear components,high temperature and dust or other harsh environment making the measurement doesn’t work,the modeling and control of the iron making process of blast furnace has been a difficult problem in the field of metallurgical engineering and automatic control.In addition,the temperature,Si content,S content,P content and other parameters of the molten iron is difficult to use the existing detection methods for real-time detection,and the off-line analysis has a long lag time(more than an hour).In actual production,because of the key information of molten iron quality can not be timely feedback,directly leading to the daily production operation of the blast furnace is not timely,and seriously restricting the quality control and optimization of the blast furnace to achieve.Therefore,in order to achieve the closed-loop optimization of hot metal quality parameters,we must first establish an effective multi-parameter prediction model of hot metal quality parameters.Due to the complex dynamic characteristics of blast furnace ironmaking process,it is difficult to establish the mechanism of multiple quality of molten iron(white box)model,and data-driven black box modeling is an effective method to solve the problem.The existing data driven methods for the modeling of molten iron quality parameters,such as BP neural network and support vector machine,generally have the problem of slow convergence speed,poor generalization ability,and poor practicality and so on.The new Random Vector Functional-link Networks(RVFLNs for short)algorithm has the characteristics of fast speed,generalization performance and suitable for multi-output modeling.It has been widely concerned and has been used in the modeling of blast furnace hot metal quality parameters.However,in the process of ironmaking,there are often outliers in the measured data due to the failure of the instrumentation and the transmitter and other abnormal interference.Since the output weights of the traditional RVFLNs are derived from the least squares,the robustness of it is insufficient and it is easy to be disturbed by the outliers,resulting in the model being inaccurate.To solve these problems,this thesis carries out research on multivariate molten iron parameters quality modeling based on RVFLNs and conduct experiment on a 2600m3 BF of Liuzhou Iron and Steel Company in Guangxi province with the support of National Natural Science Foundations,"High-performance operation control and implement technology of large blast furnace"(61290323)and "Experimental verification platform construction and application verification of large blast furnace high-performance operation control"(61290321).The main contributions are given as follows:(1)Firstly,we determines four quality parameters(the temperature of molten iron,Si content,P content,S content)which need to be modeled by analyzing the importance of the modeling of the quality of molten iron in the process of the blast furnace.Blast furnace is a typical multi variable system,which has a large influence on the quality of molten iron,and there is a significant correlation between the variables.Combined with the analysis of the mechanism of blast furnace,the current situation of detection and canonical correlation analysis,main variables that affect the quality of hot metal are selected as the model input variables from many related variables to reduce the dimension and difficulty of modeling.These selected variables are bosh gas volume,flow rate of cold air,flow rate of rich oxygen,gas permeability,oxygen enrichment rate and burning temperature.(2)In order to overcome the shortcomings of traditional RVFLNs,this paper improves the robustness of RVFLNs based on the theory of robust estimation to overcome the influence of outliers on the model.In order to further enhance the generalization ability of the model,structural risk is introduced into the algorithm.Most of the robust estimators are based on experience to select the weighting function and determine the relevant harmonic parameters,which are inefficient and have poor precision.In this paper,a Cauchy function weighting method is proposed from the point of the distribution of errors,and the related parameters are determined by analyzing the distribution of errors.At the same time,according to the dynamic characteristics of the blast furnace ironmaking process,the output self-feedback structure is adopted.In detail,it means that the input and output of the former time are introduced into the input layer of the model to establish the order model with dynamic characteristics to improve the prediction accuracy of the model.The industrial experiment and comparative study of the proposed method are carried out in a 2600m3 BF of Liuzhou Iron and Steel Company.The results show that the proposed method has higher accuracy and stronger robustness than the conventional method.The proposed method can effectively overcome the interference of the outliers in the data,and realize the high-precision modeling and prediction of the hot metal quality parameters under the condition that the data outliers are difficult to avoid.The proposed method plays an important role in guiding the operation of blast furnace.(3)In view of the strong time variation characteristics of the process of blast furnace ironmaking,the online sequential learning technique is introduced to improve the adaptability and practicability of the model.Based on the proposed Cauchy distribution weighting M-estimator RVFLNs and online sequential learning algorithm,we propose a Robust Online Sequential Learning RVFLNs(Robust OS-RVFLNs for short)model.The online sequential learning method enables RVFLNs to have the ability to process large amounts of data,and the algorithm itself has good nonlinear generalization ability and fast learning rate to ensure the feasibility of online modeling.At the same time,to eliminate the "data saturation" phenomenon with the increasing of data and improve the accuracy of the model,we use the forgetting factor method to limit the influence of historical data and expand the effect of current data.At the end,industrial test and comparative study based on the proposed method are made on the 2600m3 blast furnace of Liuzhou Iron and Steel Company.The results show that,the proposed method can effectively overcome the influence of the variation of blast furnace operating conditions on the robust modeling performance.When the operating conditions change,the proposed method can update the model parameters online through the online sequential learning mechanism,which effectively improves the model’s self-adaptability and ensures the accuracy and stability of the on-line estimation of molten iron quality parameters. |
PDF Full Text Request