Prediction for the development data of oil field with multi-variable phase space reconstruction method and support vector machines

Abstract

Time series analysis is a branch of the strong application of statistical probability. It has a wide range of applications in the field of industrial automation, hydrology, geology, meteorology and other natural domain. However, the application in the oil field development is not extensive. Currently the one-dimensional single variable time series analysis method is used to predict oil and water production. This method, however, is completely isolated without considering the relationship between oil production, water production and pressure. Moreover, it does not take advantage of the evolution and essential characteristics of the entire reservoir system. In this paper, we use multi-variable phase space reconstruction method, not only considering the variation of historical oil production, but also taking the effect of the pressure change and water production change into consideration. This method can provide the information for each prediction and other sequences. The amount of available information had increased significantly, and the accuracy of the prediction had improved greatly.