A Deep Learning-Based Method for Predicting Lithological Sequence Models Using Seismic Data

Abstract

The invention discloses a method for predicting a lithology sequence model based on deep learning using seismic data. It includes the following steps: 1) measure the lithology data of the target section on the well in the work area and the seismic data of the side channel as training data; 2) normalize the seismic data of the side channel of the well, and convert it to -1 to 1; 3) with the well bypass seismic data and the lithology data on the well after step 2) processing, adopt the stacked cyclic neural network model and the sequence to train the sequence cyclic neural network model respectively, and use the well bypass seismic data In order to observe the data, the lithology data on the well is the target data, and iterative calculation makes the learning model converge; 4) Apply the learning model calculated in step 3) and input the actual seismic data to obtain the predicted lithology sequence. The invention can generate the lithology data body which can effectively reflect the reservoir distribution under the control of the seismic data sequence, solve the reservoir prediction problem between wells, and provide the basis for exploration and development.

Description

technical field [0001] The invention relates to a deep learning-based method for predicting a lithology sequence model by using seismic data, belonging to the field of reservoir prediction for petroleum exploration and development. Background technique [0002] Lithology prediction is one of the important means of reservoir prediction. Reasonable lithology prediction results are helpful to carry out the analysis of sedimentary facies distribution and sedimentary evolution law, and then predict the spatial distribution of favorable reservoirs to guide exploration and development deployment. So far, the research on the identification and prediction methods of lithology has mainly focused on the relationship between well logging curves and lithology sequences, while less research has been done on the relationship between seismic waveforms and lithology sequences. On the one hand, it is limited by the limitations of available methods. The use of seismic data to predict lithology...

AI Technical Summary

Problems solved by technology

Under such conditions, conventional forecasting methods are difficult to deal with the problem of variable-length sequence input and variable-length sequence prediction

Images