The Research Of Steel Price In Domestic Market Forecasting Model Based On CEEMDAN_GA_KELM

Steel has not only become an important industrial infrastructure for ensuring the smooth operation of all sectors of the national economy,but its price fluctuations have increasingly become an unstable factor that constrains the smooth operation of countries' economies.In 2017,China became the country with the largest steel exports in the world,and its import volume reached the eleventh in the world,which had a certain impact on the stable operation of the Chinese economy.In the unpredictable international steel market,steel is the main molded product of steel and the most important bearer of steel prices.If we can correctly predict the price trend of steel,we can prevent price fluctuations in advance and reduce the possibility of occurrence.The risk of maintaining its own interests.At the same time,correctly predicting the price of steel in China can also enable Chinese steel mills,real estate,automobile manufacturers and other enterprises that need to use steel to respond to changes in market prices and supply and demand in a timely manner.Therefore,paying close attention to China's steel market,exploring the potential reasons for steel price changes,and making a reasonable forecast of price trends are of great significance to both countries and enterprises.As an important industrial basic material,steel has many attributes such as politics,commodities and finance.This paper takes the monthly data of hot rolled coil and cold rolled coil as the research object,considering not only the development law of price time series itself,but also Grey correlation analysis(GRA)screened seven important factors affecting price and proposed the CEEMDAN_GA_KELM steel price forecasting model.In this paper,the following two aspects are mainly studied: firstly,the ultimate learning machine(ELM)is improved by learning gaussian kernel function principle of SVM,and KELM is obtained.Therefore,KELM model also needs to select nuclear parameters and penalty coefficients.In this paper,GA_KELM model was established and genetic algorithm(GA)was used to optimize the nuclear parameters and penalty coefficients of KELM.The empirical analysis of monthly price of hot rolled coil shows that the mean square error of GA_KELM prediction is 69181,which is 21.16% and 13.50% higher than that of KELM model and SVM model.Secondly,the paper will be mainly used for adaptive noise signal technology the complete set of empirical mode decomposition(CEEMDAN)using in the study of nonlinear nonstationary sequence of prices,steel price sequence is decomposed into different frequency components,each component using GA_KELM model to forecast,will get the final weight predicted results combined reconstruction of steel price forecast.The empirical results show that the relative error of CEEMDAN_GA_KELM prediction is 11.7702%,and the prediction accuracy is 4.62% higher than that of GA_KELM model alone,which indicates that the prediction method used in this paper is feasible and effective.On the basis of this model,ARIMA model was used to predict the factors linearly,and CEEMDAN_GA_KELM model was combined to predict the hot rolled coil price in the next 12 months and put forward corresponding Suggestions.