Prediction of reservoir saturation field in high water cut stage by bore-ground electromagnetic method based on machine learning

Abstract

The prediction of oil saturation by bore-ground electromagnetic method provides a new expression for reservoir fluid. However, due to the complexity of the internal environment of reservoir and the influence of construction cost, the signal excitation point often cannot cover the whole vertical reservoir range. In this paper, based on the results of single-layer saturation field distribution collected and processed by bore-ground electromagnetic method, six machine learning algorithms, decision-tree, random-forest, k-nearest neighbor, AdaBoost-decision tree, AdaBoost-random forest and AdaBoost-k-nearest neighbor, are established by using machine learning algorithm to introduce inter layer reservoir flow correlation parameters. The fitting of prediction model is detected by learning curve in machine learning algorithm, and the hyper-parameters of the model are optimized by grid search, finally a machine learning algorithm suitable for predicting the saturation distribution by bore-ground electromagnetic method is evaluated and the remaining oil saturation of other layers in the vertical direction is predicted. The results show that the new method can effectively overcome the defects of the bore-ground electromagnetic method, the vertical saturation field of each layer can be reasonably predicted, and the prediction accuracy is more than 90% when it is applied to the prediction of saturation field in block 2 of Gangdong oilfield. The oil saturation distribution results are more similar to the actual reservoir.