Fracture density prediction using CNN-LSTM deep neural network for geologically complex geothermal reservoirs

Abstract

Summary We designed a CNN-LSTM deep neural network to characterize fracture density from seismic data for identifying upflow locations in geologically complex geothermal reservoirs. The proposed method utilizes a convolutional neural network (CNN) encoder and two long short-term memory (LSTM) networks to calculate fracture density based on seismic attributes and formation micro-imaging (FMI) logs. According to inversion results on 3D seismic data, the CNN-LSTM deep neural network predicted fracture density more accurately on a test dataset, as confirmed by the fault likelihood attribute and fracture density based on FMI logs. The proposed workflow will provide important guidance for the deployment of geothermal wells in structurally complex settings.