The Method Of Identification Of Water-flooded Based On Space Transformation

The identification of reservoir water flooding Layer is a crucial problem which is urgent to be solved during the mid-later period of water drive oilfields development.Since long-term water-flooding changes reservoir property in middle or late period of oil field development, the amplitude and shape of logs will show some changes correspondingly. A pattern recognition technology describing the amplitude and shapes of logs is used to improve recognition accuracy for the grades of watered-flooded.First the paper discusses the research status of flooded layer identification, on the basis of tackling the mapping relation between civic signals in logging in water flooded layer and the flooded grades, using this mapping to predict the level of submerged corresponding to a new water-flooded zone well-logging data is also studied. Since the data samples of water-out reservoir are finite, this paper applies support vector machine theory which is the most beneficial instruments in solving the problem of classification in the situation of small example scale swiftly develop in recent years. With a comprehensive statistical learning theory as a base, support vector machine overcoming the " dimension disaster " , " over learning" and other problems appeared on traditional theory that are still not solved .It is a kind of way to effectively improve extension and generalization ability of learning machine.Finally, aiming at the difficulties of how to extract logging signals features , a feature extraction method (Space Transformation–based Feature Extraction)that has possess more universality is put forward based on the existing methods. It is proved that the space transformation satisfy orthonormal property can withdraw the signal characteristic and ensure the sample information with a little loss or no loss. At present spatial transformation has been already successfully used in other fields of pattern recognition but not been widely used in flooded layer identification, the paper applies the space transformation theory to identification of water-flooded layer. 350 samples decimated from 4 core wells is conducted a lot of experiments and the results show that the correct rate of testing samples is up to 78.1% and this method have high recognition rate and good generalization property.