Application Of Water-flooded Zone Identifying Based On Computational Learning

Today all of the big oilfields in our country comed into the later-period of exploration, and water-injection becomes an important method of oil-field exploitation, the recognition and appraisal technology of water-flooded became an important step in reservoirs explanation for describing the oil-deposit more precisely, finding out the water-flooded patterns,reducing the oil-extraction cost, improving the quantity and quality of extracted oil, enhancing the whole economic efficiency of oil-field.Firstly, the status quo of water-flooded identifying were investigated, including three aspects that are laboratory geology analysis technology, global chemical technology and global physical technology,rapidly-developing water-flooded identifying automatic recognition technology were concluded,as well as the features and difficulties.Secondly, the basic theory and knowledge were investigated. Starting with the the theoretical background analysizing, PAC learning model,the foundation and core of Computational Learning Theory, and its expansive model,VC dimension and sample complexity that standing for the volume and complexity of learning system were also investigated, PAC-learnability of some concept class were analyzed and listed.We concluded that as a formal frame analyzing and designing algorithm, the research achievements of computational learning theory provides important guidance.At last,Boosting learning and SVM which are developed on Computational Learning Theory were investigated to improve the classification ability and generalization ability.For SVM flaw,by combining Boosting algorithm and SVM,ensemble SVM is investigated,a improved SVM algorithm named BoostingSVM is proposed and used in identifying the water-flooded layer, that a new method to identify water-flooded layer automatically is formed——water-flooded zone Identification based on computational learning theory. The experiment results demonstrate that BoostingSVM algorithm based on computational learning theory is an effective one on automatic water-flooded identification and classification , and is better than original SVM algorithm on the key index such as classification ability and generalization ability.