Spatial Support Vector Machine Model For Comprehensive Evaluation Of Oil And Natural Gas Reservoirs

Oil and natural gas is the lifeline of the national economy,which has a major impact on the development of national economy and improvement of people's quality of life.In order to meet the ever-growing demand of oil and natural gas,on the one hand,we need to reasonably,fully tap potential of oil field which has been developed,on the other hand,we need to improve current existing methods to finding new oil fields.So comprehensive evaluation of reservoir is the key step to find the distribution of oil and natural gas resources.Because of the complexity of the geological process,the distribution of oil and gas reservoirs in the space is heterogeneous,and the high difficulty and high cost characteristics of drilling implementation,drilling is always preferred to be designed in the favorable reservoir positions,which makes the geological data inhomogeneous in spatial distribution.Reservoir objects are the reservoir phenomena which describe the specific geological space.The spatial characteristics and distribution rules of the reservoir geological bodies lead to a certain spatial distribution pattern of the reservoir object dominant or recessive.Moreover,the spatial patterns and attributes of the reservoir objects vary with the different research scales and levels,and there are some spatial or spatial dependence among the reservoir objects.This paper focuses on the data of oil and gas drilling data with a small amount,uneven distribution of different types,and uneven distribution of spatial distribution.Without changing the number of samples and their spatial distribution,and without the knowledge and experience of the field of oil and gas,the paper systematically studies support vector machines,voronoi diagram and spatial adjacent index,and builds an integrated expression model of reservoir evaluation object attributes and neighborhood index features,combined with analytic hierarchy process method based on the k-order adjacency index in Voronoi diagram mode,we extract spatial adjacency index from oil and gas drilling data and combine it with drilling attribute characteristics.Then,we combine G-Mean evaluation indicators,grid search,stratified sampling,and k-fold cross validation to build a spatial support vector machine model.The paper designed and developed support vector machine program,and carried out the experiment in the eastern region of the Sulige gas field in the Ordos Basin.The experimental result shows that compared with the classical support vector machine,the spatial support vector model has improved by approximately 12.1% for class I classification accuracy,6.13% for class II,reduced by 0.9% for class III,and improved by 1.2% for total classification accuracy.After the experiment,the following conclusions can be obtained.1)Facing with the drilling data,where the amount of data is small and the sample count about classes is imbalanced and the spatial distribution is uniformity,without changing the number of samples and their spatial distribution,compared with the classical support vector machine,the spatial support vector machine proposed in this paper can get a higher G-Mean evaluation index and can improve the classification accuracy of class I and class II and the total classification accuracy.2)Without the knowledge and experience of the field of oil and natural gas,the spatial support vector machine model can obtain a better classification accuracy on class I,class II and the whole,which can be used for comprehensive evaluation of oil and natural gas reservoirs.