Identification Of Blast Furnace Condition Based On FLS And Its Application In Temperature Prediction

Iron and steel industry is the lifeblood of China's national economy,and blast furnace iron making as its core process has a direct impact on the development of steel-making industry and energy saving,so research on blast furnace iron making technology has important research significance.Due to the long time lag of the blast furnace,the smelting process of the blast furnace is not easy to be controlled.Therefore,the smelting level of the blast furnace can be greatly improved if the blast furnace temperature can be accurately predicted so that the operators can make timely adjustments to strategies according to the future forecast situation.The direct measurement of furnace temperature is costly and difficult,while the silicon content of molten iron can not only represent the furnace temperature well,but also reduce the measurement cost greatly.As a result,this thesis tries to establish a silicon prediction model to indirectly predict the furnace temperature through silicon content.The current furnace temperature prediction model doesn't evaluate the internal state of the blast furnace,which makes the model lack a reliable basis of judgment when selecting training data.This is mainly reflected in the fact that the selected training data contains a variety of state data or does not contain data in the same state as the predicted sample.which will reduce the prediction accuracy of the model.In order to improve the prediction accuracy of the furnace temperature model,the following researches are carried out in this thesis:(1)Conduct the research on state identification related algorithms,propose a blast furnace state identification based on the flexible least square method.On this basis,the data with the same state as the predicted sample is selected as the training set,making the selection of training data more accurate and reliable.(2)Establish PCR model and SVR model to fitting analysis the blast furnace data,select the SVR model suitable for blast furnace data to predict the silicon content of molten iron,and establish multiple SVR models based on the different selection methods of training data,and compare the prediction effects of each model.The results shows that the prediction accuracy of the model established after state identification has been improved compared with other models.It proves the necessity and feasibility of dividing the internal state of the blast furnace,and also illustrates the effectiveness of the flexible least square method for identifying the internal the internal state of the blast furnace.(3)Aiming at the blast furnace model error,a multi-model prediction interval counting method is proposed,which further improve the overall prediction accuracy of the model.To sum up,the research program in this thesis enables the model to maintain a good prediction accuracy when dealing with changes in the internal state of the blast furnace,which has a good guiding significance of actual production operations.