PREDICTION OF CO2 GAS SATURATION DISTRIBUTION BASED ON DEEP LEARNING USING DEEP NEURAL NETWORK (DNN) ALGORITHM

Abstract

Multiphase flow analysis is essential for resolving subsurface flow issues in CO₂ capture and storage (CCS) systems. Predicting the distribution of CO₂ gas saturation is one example that is quite useful for evaluating multiphase flow. Multiphase flow simulation is typically performed using numerical simulations, such as the TOUGH2 simulator. Ordinary numerical simulations, on the other hand, have some limitations, such as high grid spatial resolution and significant processing costs. One option for estimating the distribution of CO₂ gas saturation is to employ deep learning with specific algorithms. A deep neural network (DNN) is a highly effective deep learning approach. A deep neural network is a network structure made up of three interconnected layers: input, hidden, and output. DNN learns from the input data about the previously constructed architecture. As input, DNN requires a significant amount of data train. The trained DNN model is then used to automatically estimate the distribution of CO₂ gas saturation. This algorithm is capable of dealing with complex data patterns, particularly gas saturation in multiphase flow issues. The reconstruction loss results revealed that the loss value lowers as the number of epochs grows. Furthermore, the model with 5 epochs and 0.001 regularization weight had the least error value 0.43. As a result, while this model is adequate for predicting the distribution of CO2 gas saturation, additional research is required to achieve more ideal outcomes.