Abstract: Firstly, a comprehensive analysis of the key factors affecting consumer demand for gasoline and diesel is made for selfrelevance, complexity and data volume, etc. A principal component analysis is made to reduce the dimension of the sample data, and a new set of samples is formed. Then, by improving the support vector machine model and introducing a dynamic factor in the timing of its foundation, and the demand for gasoline and diesel last year historical data into the model as the timing of the feedback factor, thus forming a new dynamic feedback fitting model, an appropriate demand forecasting model is established. Finally, a case study is made on forecasting demand for gasoline and diesel in the 1996~2012, and the proposed method of predicting and gray GM (1,1) model, and BP neural network model are analyzed. The results show that the improved prediction method relative to the GM support vector machine principal component analysis (1,1) model of the prediction errors are respectively 72.7%, 74.86% lower, and that comparing with the BP neural network, the prediction errors are reduced on average by 81.3%,81.66%. Results show that the principal component analysis using improved support vector machine method is superior to existing methods, which proves the effectiveness and superiority of this method.