Study On Chaotic Time Series Prediction Of Petroleum Futures Prices

As a sensitive factor, petroleum market affects the world economy and politics status, which influence petroleum market meanwhile. Petroleum futures market is a product of stock market which developing to a certain level. It's a very complex and opening nonlinear dynamic system. Petroleum futures market has the function of discovering price and avoiding risk, however, it brings some potential risk to the users. Price risk is one of the core factors in petroleum futures market risk. This paper emphasizes that"phenomenological-only analysis"is more feasible than"mechanism-only analysis"to the petroleum futures market based on the qualitative analysis of price influencing factors and the quantitative analysis of chaotic characteristic. This paper also brings forward that the petroleum futures market price time series is a chaotic time series through the phase space restructuring process and analysis the chaotic characters. The Prediction of the petroleum futures market price time series can be a prediction of the chaotic time series. This paper is based on the partial neighboring field prediction of the chaotic time series prediction, using the embedding dimension of phase space restructuring as the division length of petroleum futures market price time series and divide it to several sub-series. The number of sub-series is the number of phase points in the Phase space. Some phase points are selected to compose the input sample set of the Artificial Neural Network through the cluster analysis and carry out the prediction process. In the selection process of the partial neighboring phase points, this research improves the correlation function compared to the method of using the interrelatedness, a general relevant measure method, as the relevant function of phase points in former researches. It also combines the interrelatedness and correlation coefficient with the horizontal drift and scope expansion for the first time, such a method can be more feasible to reflect the preference when keep the universality of phase points relevant measurement. From another point of view, using the characteristic property of various components of correlation function to represent the relation of phase points, the amelioration can be described as a characteristic property extraction process. Not only the relation of phase points is reflected better, but also the cluster analysis can be developed on this. This paper chooses the hierarchical cluster based on the analysis of the phase point relations and the various cluster method. Some historical phase points are found through the cluster analysis, which belongs to the same set with the prediction phase points. Furthermore, this paper constructs the partial neighboring field needed by the price chaotic time series prediction. The general regression Artificial Neural Network is used to receive the relation of predicted phase point and its partial neighboring field, also separate the predicted value. The Radial Basis Function (RBF) and General Regression Neural Network (GRNN) are selected because of its excellent character of approximating, categorization and arithmetic speed compare to BP Neural Network , its learning algorithm can be with teacher or non-teacher. General RBF network uses the K-means method, in this paper, the relevant measure of phase point and the hierarchical cluster analyze can play a role as confirming the partial neighboring field in phase space composed by chaotic time series, also makes an improvement on learning algorithm of RBF neural network. To test and verify the modeling research described above, the demonstration analysis is put forward in the final part of the paper, a software system is developed based on the prediction model, a result can be find out that the improved prediction model is more accurate than traditional models, the predict precision has been advanced to a certain degree.