Abstract: A good command of the price changes of thermal coal is closely related to the pricing strategy of companies participating in the power market and the formulation of policies to optimize internal operating costs. To solve the problem that there are a large number of influencing factors leading to the change of the coal price and frequent changes of the influence weight,this paper establishes the Least Square Support Vector Regression（LSSVR）model of the coal price in the short and medium term on the basis of the reconstruction of feature space by means of refinement analysis,factor checking and seasonal difference and other methods of influencing factors of the coal price. Firstly,according to econometrics and other theories,the influencing factors of the coal price are screened,and the main influencing factors are extracted by co-integration check and Granger test to compress the characteristic factor dimension. By establishing the data set of multi-year synchronous comparison timing sequence information and using seasonal difference to eliminate the influence of periodic factors and random interference in timing sequence information,the reconstruction of feature space is realized and the quality of input data is improved. Furthermore,a trend extraction learning model based on LSSVR is established,and combined with periodic prices and residuals,the medium and short-term coal price forecasts are realized. Finally,several comparison models are constructed to verify the effectiveness of the proposed model.