A carregar...
Publicação
Transmissibility Prediction: A Deep Learning Approach
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.91 MB | Adobe PDF |
Autores
Orientador(es)
Resumo(s)
Transmissibility is a very important issue in the study of fractured rocks, as it is directly related to the efficiency of drilling and extracting oil wells, water reservoirs, and even gas exploration. In this piece of work, based on data from transmissibility simulations performed in oil fields in Norway, three different machine learning approaches were applied for predicting the transmissibility of fractured rock areas. First, the fracture diagram image was applied in two different Neural Networks architectures: GoogleNet and ResNet. Second, from the fracture diagram image, it was performed a decomposition of all fracture lines (scratches) on each image into X-axis and Y-axis and it was sent to the same two Neural Network architectures on the previous approach (GoogleNet and ResNet). And finally, it was performed a discretizing continuous variable, and applied on neural network ResNet, thus performing a multi-class classification for predictions instead of regression. Overall, this study provides contributions for transmissibility prediction on oil well fields. Creating options to the traditional technique of calculating transmissibility by computer simulation which is very costly and time-consuming.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics
Palavras-chave
Machine Learning Neural Network Deep Learning Oil Water Rock Fracture Transmissibility