Novel Approach To Predict Potentiality of Enhanced Oil Recovery

摘要

The potentiality prediction of enhanced oil recovery (EOR)rnis the basis of EOR potentiality analysis as well as the robustrnguarantee of the reliability of analysis results. In the light ofrnstatistical learning theory, establishing an EOR predictivernmodel substantially falls within the problem of functionrnapproximation. According to Vapnik's structural riskrnminimization principle, one should improve the generalizationrnability of learning machine, that is, a small error from anrneffective training set can guarantee a small error for therncorresponding independent testing set. The up-to-date resultsrnfrom studies on statistical theory in recent decades even recentrnyears are firstly applied to EOR potentiality analysis. Thernapplications of group method of data handling (GMDH),rnImproved BP artificial neural network, and support vectorrnmachine (SVM) are discussed. The comparison of the resultsrnfrom three methods indicates that SVM can pay more attentionrnto both the universality and extendibility of a model when thernsamples are very limited, which shows a good prospect of itsrnapplication. A method used to generate a set of samplerntheoretically is developed in this research by combiningrnquadrate designing, reservoir simulation, and economicalrnevaluation.