The Research Of Compressed Sensing Based International Crude OIL Price Forecasting Approach

International crude oil price has been fluctuating since 1970s. It is not only influenced by the basic supply and demand, but also influenced by the synthesis impact of many other factors including weather, stock level, economic growth, political factors even unconventional events. The impacts of multi-factors lead to the complex characteristics of non-linearity, non-stationary, seasonality and irregularity of crude oil price. Under this condition, this paper builds a novel compressed sensing based artificial intelligent forecasting approach to predict crude oil price.In detail, the approach introduces two compressed sensing based techniques including compressed sensing based denoising approach and sparse decomposition approach as the preprocess of crude oil price series, then respectively build two kinds of model based on these two data processing techniques. One model is compressed sensing denoising based artificial intelligent forecasting model, the other is sparse decomposition based decomposition and ensemble forecasting model.The compressed sensing denoising based artificial intelligent forecasting model is based on the concept of denoising based forecasting, first the compressed sensing denoising approach is applied to eliminate the noise of crude oil price series in order to decrease the impact of modeling performance of artificial intelligent forecasting model. Then the intelligent forecasting algorithm is used to model and forecast the denoised data.The sparse decomposition based decomposition and ensemble forecasting model is based on the concept of decomposition and ensemble, first according to the various characteristics presented by crude oil price series, an over complete dictionary is constructed and the crude oil price series is decomposed into components of different characteristics. Then the feed-forward neural networks are used to model and forecast each of the components. Finally the forecasting results are integrated to the final forecasting result.This paper respectively use the daily and monthly data of WTI price series for the two compressed sensing based forecasting models to conduct empirical analysis. And some conclusions are given:on the one hand, the compressed sensing based two forecasting models both achieve the highest forecasting accuracies, indicating the effectiveness of the compressed sensing based forecasting approach; on the other hand, the compressed sensing based two forecasting models both achieve the highest forecasting accuracies, indicating the robustness of the proposed approach. Moreover, the empirical results also indicate that this approach is especially effective in forecasting time series of complex non-linear characteristics.