The Research And Application Of Neural Network Pattern Recognition Technique For Oil And Gas Geochemistry Base On MATLAB

The geochemical exploration of oil and gas is a kind of economic and rapid method of exploration to search petroleum resources, which finds every trace geochemistry index abnormity in the superficial medium caused by the hydrocarbon microseepage of deep underground hydrocarbon reservoir with the aid of all kinds of detection means. The traditional route is mainly to get background and abnormal value and discover the range and position of abnormity by the multiple statistical analysis of every geochemical exploration index and their relationship in advance, then analyze integratively the abnormity of every index, and finally forecast the hydrocarbon reservoirs by combining the geological and geophysics data. It is very mature in application of oil and gas geochemistry. But such method has multi-solutions. So this paper tries to adopt the neural network pattern recognition nonlinear technique, which unifies geology, geochemistry and geophysics to a neural network model from another angle, and then forecast the hydrocarbon reservoir. This method has the advantages of self-learning, self-organization , self-adaptability, robustness and generalization, which can solve complicated nonlinear problems. And neural network has a variety of applications in exploration of geology,gravity, magnetism, electricity, and seismic.This paper relies on the item of the study of geochemistry information recognition and forecasting technique in complicated oil-gas field, which is included in the groping topic of the study of united inversion of high accuracy heavy-magnetism and electric shock in the national 863 plan of resources and circumstance field.Firstly, this paper discusses the actuality of the neural network in the oil-gas exploration field. Secondly, it introduces particularly the inner design of neural network based on MATLAB, and concludes the principals, application areas and parameter setting of some kinds of advanced algorithmic BP neural network. And then, it establishes a accurate BP model to classify and forecast the 840 samples, and makes integrated interpretation and estimate to the hydrocarbon reservoir in this area, which is based on the geochemical exploration data of huagou area in shengli oil-gas field, through the sorts of 43 wells known, it adopts gradient descent momentum algorithm and determines reasonable parameters by learning-training time after time. Finally, it makes contrastive analysis between the intrinsic geochemical exploration results and the prediction results caused by neural network.Based on the prediction results, the 840 samples are classified into five: methane, carbon dioxide, combined shaft, common and prediction failure. Because prediction failure accounts for a small amount, and it almost have no influence to the results. It can be seen that the neural network model established in the paper is reasonable and accurate, and the prediction results nearly agree with the intrinsic geological materials, also, several unknown areas are predicted, which supplements intrinsic materials perfectly. In addition, aiming at these areas, it gives emphasis to the carbon dioxide and combined shaft areas. So the model is available in other analogous areas for prediction.