The Research Of Rebar Futures Price Forecast Based On PSO-GPR Model

With the rapid development of China's economy,urbanization and infrastructure construction,steel industry has become a pillar industry in China,which plays an irreplaceable role in the national development.Rebar is the main engineering material of reinforced concrete structure,the demand and output has been maintained at a high level.However,because the price of steel is affected by the demand change of steel enterprises and the price fluctuation of iron ore,many steel production enterprises and circulation enterprises are facing greater risks.In 2009,rebar futures were listed.Because the futures market has the function of price discovery and risk aversion,many iron and steel enterprises and individual investors actively participate in rebar futures trading to avoid price risk through hedging,or to obtain profits through futures market speculation.Therefore,the statistical analysis of rebar futures price is of great significance.In this paper,particle swarm optimization based Gaussian process regression model(PSO-GPR)is used to predict the futures price,and the prediction effect is compared with particle swarm optimization based support vector regression model(PSO-SVR).The main work includes:(1)from the perspective of weight space and function space,the expressions of mean vector and covariance matrix of prediction values based on Gaussian process regression model and their relationship with kernel function are derived respectively;(2)the data of spiral steel futures with contract code rb888 in recent 10 years are collected,and the SVR models based on different kernel functions are analyzed by particle swarm optimization algorithm And Gaussian process regression model to find the optimal parameters.(3)The trained PSO-GPR model is used to fit the prediction data.Compared with the prediction effect of PSO-SVR model,it is found that the Gaussian process regression model based on particle swarm optimization algorithm can achieve better and more stable prediction effect,and the GPR model based on Gaussian kernel function performs the best,which can provide some reference for the screw thread steel futures price prediction.