Blast Furnace Temperature Fluctuations Analysis That Based On Nonlinear Additive Model

As one of pillar industries of national GDP, Iron and steel metallurgical industry plays a vital role in the economic development of our country. As the upstream process in the main body of steel manufacturing, blast furnace iron-making is the important constitute part of the iron and steel industry,it plays a very important role no matter for the development of the industry, or for energy saving and emission reduction. Blast furnace temperature is an important index of blast furnace condition identifying, by controlling the furnace temperature can ensure the smooth progress of blast furnace condition. The hot metal silicon content, has been used as a characterization of furnace thermal state indicator for long time. Therefore, to establish a reliable and accurate forecasting model,used for blast furnace ironmaking personnel guidance for furnace temperature control, not only to theoretical researching, but also to the production practice of iron and steel industry, is a great guiding significance. However, the blast furnace ironmaking process is very complicated, its operation mechanism usually has nonlinear, high dimension,uncertainty, and great noise, how to overcome these characteristics, the main difficult problem of the model development is how to research high precision prediction model.In this paper, on the basis of predecessors' research on blast furnace temperature prediction, the introduction of nonlinear additive model and time series method as the main research tools, collected the blast furnace production data of Baotou iron and steel group company online as the raw data by using blast furnace expert system. By a large amount of data processing,analysis and model fitting inspection,set up the blast furnace temperature prediction model based on nonlinear additive model.Additive models is a new empirical method that used in foreign countries in the 90's.By the continuous developing, it formed a certain framework in theory. At present, the additi-ve model is widely used in medicine, biology, finance, statistics, and other fields, its biggest advantage is driven by data rather than the driven by model, is not confined to an assumed form or a curve shape, it only requires the prediction variables effect can be added. At the same time, the additive model can avoid"dimension disaster", therefore, the problems of making process of blast furnace temperature data, and setting up the temperature prediction model are simplified.Firstly this paper summarizes the basic principle and process of blast furnace ironmaking, the blast furnace expert system and blast furnace prediction model at home and abroad. Then the current development of blast furnace was introduced, and gives the basic theory of nonlinear additive models and time series model, finally establishes the blast furnace temperature prediction model based on nonlinear additive models.In this paper, the sample size is 300 sets of blast furnace data collected online.Firstly, this article takes the normalized data preprocessing, and uses all the index data that collected from blast furnace inspection system to do a collinearity diagnosis.Then determines the fitting index in additive model, finally uses R software by GAM package to fitting model, through the P value, choose the optimal model, and uses data to check it.This article uses the time series AR model and nonlinear additive models to forecast .The forecast results are compared, the results show that the prediction model based on nonlinear additive model is better than the original time series AR model.Nonlinear additive model improves the hit ratio significantly,and have a better prediction effect, which proved that the nonlinear additive model has practical value in the blast furnace temperature predicted.