Research And Application On Prediction Technique Of Fracture And Cave Based On Support Vector Machine

Fracture and cave prediction, a world-class issue, is the focus of oil and gas exploration and development. The accuracy of prediction on fracture-cave development degree and location will have direct impact on the effect of the exploration and development, so it has always been attracted great importance by the oil-gas exploration experts and workers. Therefore it is a key procedure to seek a better method for fracture-cave reservoir prediction. From 1970s, many scholars at home and abroad have been devoted themself to the study of fracture-cave systems prediction. They put forward plenty of methods, such as pattern recognition, neural networks, grey genetic algorithm, and so on. It is true that these methods have achieved certain degree of success in the theory and practical applications, but there are still a lot of problems, for example,limited using of seismic attributes, lacking of quantitative description, low prediction accuracy.In addition, the use of seismic data to predict fracture-cave reservoir no matter what method has been chosen, large number of seismic attributes must be extracted and classified. Added with the uncertainty and randomness of geology itself, So, it certainly involves the preferential optimization of seismic attributes.In response to above problems, this paper proposes a fracture-cave reservoir prediction method relying on support vector machine. The foundamental idea is: firstly, using rough set to optimize the seismic attributes and obtaining combination of attribute reduction; then, using seismic attributes which have been reduced as condtional attribute, fracture-cave development degree as decision attribute to build fracture-cave reservoir prediction model basing on support vector machine. Finally, the model was applied to the Kashan block in middle Iran basin to test the practical applicational effects of the method.Through researching, the thesis has made the following progresses:1. Using rough set to reduce the seismic attributes can extract the main characteristic parameters that can reflect the development degree of fracture-cave reservoir without changing the original classification capacity. This procedure also greatly brings down the dimension of attributes, reduces the calculation, simplifies the model and makes the model more reliable.2. Combining rough set with support vector machine, using rough sets as the pre-system and support vector machine as the rear system, building a fracture-cave reservoir prediction model basing on support vector machine. The model has high rate of return to contractors and accuracy of identification and prediction.3. Applying the model this paper built to Qom Formation in the Kashan Block, achieved the horizontal prediction in F-E,C and B-A layers, which could provide reference for the further exploration and development of this block.