Research On Prediction Method Of Total Organic Carbon In Shale Based On Machine Learning

The organic matter content is usually expressed by Total Organic Carbon(TOC),which is one of the important indicators for the evaluation of hydrocarbon generation capacity of shale reservoirs.How to accurately predict TOC using seismic data is one of the important issues to be solved in current shale oil and gas exploration.Considering the complexity of the shale reservoir itself,this paper introduces the machine learning method and uses the artificial intelligence nonlinear prediction to realize the quantitative prediction of the TOC using the logging data and the seismic prestack inversion results.The main workloads completed in this paper include:(1)Theoretical analysis of shale rock physics characteristics related to TOC,investigating existing TOC prediction methods,and discussing the advantages and disadvantages of each method.(2)Several popular machine learning algorithms were investigated.Two machine learning methods suitable for TOC prediction were selected and implemented through comparative analysis: random forest algorithm and support vector regression.The model dataset trials are performed on the two methods.The results show that the support vector regression is more accurate when the training data is limited,and the random forest algorithm is more accurate.(3)The TOC sensitive parameter analysis was carried out by using the logging results of the Wufeng-Longmaxi Formation shale in southern Anhui.(4)TOC prediction is carried out for logging data and seismic inversion results respectively: firstly,the correlation coefficient is used to optimize the parameters,and support vector regression is used to predict the TOC log curve when the core data is limited;the random forest algorithm is applied to 3D seismic prestack inversion results with limited predictive factors.The reliability of the above method was verified by comparing the inversion results with the actual log data.