Research On The Spatial Evaluation Method Of Hydrocarbon Resource Potential Based On Remote Sensing And Machine Learning

As a kind of unrenewable resources, oil and gas is the lifeblood of economic and social development. China is rich in petroleum resources. However, the proved reserves are far below the potential oil and gas resources. With the gradual development of exploration and exploitation, it is more and more complicated to geological condition and difficult to exploit the degree. So how to accurately evaluate hydrocarbon resource potential and predict oil-gas favorable exploration zone have been the crucial research topic, to increase the success and reduce the risk and cost of the oil and gas production.A lot of effective theories and methods have been put forward by domestic and foreign scholars in the spatial evaluation of hydrocarbon resource potential. Nevertheless, there are some problems in the study of evaluation factors and evaluation models, such as the factors selection and acquisition. This dissertation focuses on the applications of remote sensing techniques and machine learning in the spatial evaluation of hydrocarbon resource potential, combined with oil and gas geological theory, GIS technology, mathematical statistics, and other related disciplines and technologies based on existing researches. The main achievements and innovations are summarized as follows:1. According to the one-sidedness in the spatial evaluation factors of hydrocarbon resource potential, an assessment criteria system is built with "regional source-reservoir-caprock conditions+regional structural characteristics+oil and gas surface abnormaly information". With the system, the evaluation factors are extented from the underground geo-anomaly information to the surface abnormaly information.2. For the hydrocarbon surface anomaly information loss caused by using a single method and the shortage of results verification and analysis based on remote sensing technology, a comprehensive extraction and validation method of oil surface abnormaly information is presented by using ASTER data. First, adopting the methods of band math, principal component analysis, mixture tuned matched filtering and split-window algorithm, temperature emissivity separation algorithm, the mineral alteration information and geothermal anomaly information can be extracted respectively. The comprehensive application of different image processing methods can reduce the bias error. Then, the mineral alteration results are quantitatively evaluated with the indices of mineral recognition accuracy and mineral recognition rate by open field survey. Finally, it has been proved that the surface anomaly information indicated the oil and gas resources combined with the regional geological background and existing data.3. In view of the research status on spatial evaluation model of hydrocarbon resource potential based on machine learning, the generalized additive model (GAM) is applied to evaluate the oil and gas resources potential in space. Besides, a "weight of evidence-support vector machine" (WoE-SVM) integration model is proposed, in which the evidence layers and their weight values obtained from WoE model are input into the SVM model simultaneously, in order to make up for shortcomings of multivariate statistics, nonlinear and small sample size in WoE model, and solve the problem of the important factor interfered with other data noises in SVM model. Through drawing ROC curves of the models, the results show that the accuracy of "WoE-SVM" integration model is the highest, next is the SVM model, the third is the GAM, and the accurate of WoE model is the lowest.4. With the assessment criteria system construction, the evaluation factors extraction and quantitative selection, evaluation model establishment and quantitative test, a set of spatial evaluation methods of hydrocarbon resource potential has been formed. The results show that the spatial evaluation process of oil and gas resource potential is reliable and useful with a case study in Yulin gas field. The prospective favorable areas can provide support to the next hydrocarbon exploration and development within this area.