Application of neighborhood information-enhanced MLSTM in reservoir parameter prediction:a case study of heterogeneous carbonate reservoirs
Accurately predicting reservoir parameters such as porosity,permeability,and water saturation is a fundamental basis for reservoir fine evaluation and oil and gas exploration.Traditional methods for the prediction of reservoir parameters often rely on empirical formulas or simplified geological models that are built upon well logging data.However,these methods tend to overlook the nonlinear relationships among well logs and exhibit poor generalization ability when applied to complex reservoirs.Facing the strong heterogeneity of carbonate reservoirs and the sequential characteristics of well logs,this paper proposes a Fused Neighborhood information Multi-layer Long Short-Term Memory network(FN-MLSTM)to address the challenge of parameters prediction of carbonate reservoir.Firstly,the Principal Component Analysis(PCA)is employed to extract independent features from well logs.Then,an unsupervised clustering technique based on the K-Means algorithm and its optimization is employed to form well groups.By incorporating the resulting neighborhood information,a Multi-layer Long Short-Term Memory network(MLSTM)is built to predict the reservoir parameters.Experimental results demonstrate that the proposed model outperforms other methods such as Long Short-Term Memory network(LSTM),Deep Neural Network(DNN),eXtreme Gradient Boosting(XGBoost),and Random Forests(RF)in terms of Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Coefficient of Determination(R2)on a test set of 22 wells in a specific region.This highlights the outstanding performance of the model in parameters prediction.Moreover,ablation experiments show that the integration of neighborhood information effectively improves the predictive accuracy of the model.
Reservoir parameter predictionWell logsMulti-layer Long Short-Term Memory network(MLSTM)Principal component analysisClustering
NETL
NSTL
万方数据
