Semi-supervised Learning Methods And Applications In Facies Prediction

With the increase of energy demand for social development,especially the demand for oil resources as the main energy source has continued to rise.This has brought new challenges to the exploitation of the original oil and gas fields.This requires the geologists to fine-tune the sedimentary facies.In recent years,with the vigorous development of computers,machine learning and artificial intelligence have made great progress.This has given geologists new solutions to the study of underground oil and gas reservoirs and sedimentary facies prediction.For the traditional identification methods in the prediction of sedimentary facies,the lack of available information and the lack of knowledge of the spatial structure of the block data are used.This paper uses the semi-supervised learning method in the Block 30 of Sulige Gasfield as an example.Sedimentary facies predictions,in order to draw more detailed description of the sedimentary facies.First,data analysis is performed on seismic data and well logging data.By analyzing the data relationship and distribution structure between sedimentary facies and seismic attributes,it is shown that seismic facies data and logging data can be combined and used effectively in sedimentary facies prediction and can be effectively used.Complementary data in the prediction of sedimentary facies will be used to improve the prediction accuracy of sedimentary facies.When analyzing the data space structure of seismic attributes,it can be found that the sedimentary facies distribution is mainly affected by attributes such as effective bandwidth,instantaneous frequency,and root mean square amplitude.Then,the semi-supervised algorithm is analyzed.Because the logging data is relatively small,but the guiding role played in the prediction is very large,so the selection of parameters in the semi-supervised algorithm is very important.At the same time,because the traditional pattern recognition methods are mostly based on statistics,they often overlook the spatial structure characteristics of geological data itself.Therefore,the data space structure applicable to various semi-supervised methods is analyzed.Finally,a semi-supervised learning algorithm is introduced to predict the sedimentary facies of the block,and the differences in results obtained by different methods are studied to further analyze the spatial structure of the block.Through the analysis of the results,it was found that when selecting the appropriate parameters,the combination of seismic data and logging data can better predict the sedimentary facies,and the prediction results are more refined,and can also basically reflect the distribution of geological data in the block.