Find in Library

On the basis of the basic principles of optimization algorithms and classification algorithms, the Self-Organizing feature Map neural network (SOM) is applied to establish the predictive model of lithology for the K-Means optimized data set including core data, logging data and well tests data. Additionally, the decision tree and support vector machine are used to build the predictive model of fluid on the basis of the lithology identification. The optimization algorithms, including genetic, grid and quadratic, are adopted to optimize the important parameters of C-SVC and ν-SVC, such as C, ν and γ, so as to accurately identify the complex lithologies and multiphase fluids of complicated reservoirs. The SOM model and the decision tree and support vector machine are utilized to process four new wells in the complicated Carboniferous reservoirs of the Wucaiwan Sag, eastern Junggar Basin. The accuracy of lithology identification is 91.30%, and the accuracy of fluid identification is 95.65%. The lithologic complexity is not the main factor leading to the differences of fluids in the reservoirs. Because the complexity and nonlinearity of data set are not strong enough, the accuracy of the decision tree model is better than that of the support vector machine. Their accuracy rates are 94.31% and 86.97%, respectively. The performance of linear polynomial function is better than that of the radial basis function RBF and the neural function Sigmoid. The classification performance and generalization ability of C-SVC are stronger than that of the ν-SVC. Key words: lithology identification, fluid identification, reservoir evaluation, computational intelligence, predictive model