Reservoir Prediction Of Seismic Data Based On Radial Basis Neural Network And Locally Linear Embedding

The development of hydrocarbon exploration highlights the importance of accurate prediction of reservoir parameters.Seismic attributes contain abundant geological information of reservoirs and can comprehensively depict underground structures,lithology as well as physical properties.Consequently,seismic attribute technology is an effective method for reservoir prediction.When seismic attribute technology is conducted,various seismic attributes need to be extracted,which means that some problems exist like attribute redundancy while acquiring a large amount of reservoir information.Thus,it's necessary to perform seismic attribute optimization to simplify the calculation process and improve the prediction accuracy.In this paper,linear principal component analysis(PCA)and nonlinear locally linear embedding(LLE)are adopted to perform dimensionality reduction on seismic attributes,generate fewer numbers of new attributes that are more effective,and remove the redundant information from the original data.The optimization results of real data show that locally linear embedding possesses better optimization effects than principal component analysis due to the nonlinearity of attribute data.Due to the complex non-linear relationship between seismic attributes and reservoir parameters,the prediction accuracy and efficiency of traditional linear methods are limited.In contrast,neural networks,with good self-learning,self-adaptive and nonlinear mapping abilities,are applied to forecast reservoir parameters.This paper first introduces back propagation neural network(BPNN)and radial basis function neural network(RBFNN),and compares their prediction results using real data.It's concluded that RBFNN provides higher prediction accuracy.Secondly,PCA and LLE are utilized to optimize the input of RBFNN,respectively,and their prediction performances are compared.The results suggest that RBFNN optimized by LLE(LLE-RBFNN)owns higher prediction accuracy and efficiency.Finally,LLE-RBFNN is applied to predict the porosity distribution of the target layer.The result is well-matched with the actual drilling data,indicating the effectiveness and feasibility of the method.