Prediction and Optimization of WAG Flooding by Using LSTM Neural Network Model in Middle East Carbonate Reservoir

Abstract

Abstract Production prediction continues to play an increasingly significant role in reservoir development adjustment and optimization, especially in water-alternating-gas (WAG) flooding. As artificial intelligence continues to develop, data-driven machine learning method can establish a robust model based on massive data to clarify development risks and challenges, predict development dynamic characteristics in advance. This study gathers over 15 years actual data from targeted carbonate reservoir and establishes a robust Long Short-Term Memory (LSTM) neural network prediction model based on correlation analysis, data cleaning, feature variables selection, hyper-parameters optimization and model evaluation to forecast oil production, gas-oil ratio (GOR), and water cut (WC) of WAG flooding. In comparison to traditional reservoir numerical simulation (RNS), LSTM neural networks have a huge advantage in terms of computational efficiency and prediction accuracy. The calculation time of LSTM method is 864% less than reservoir numerical simulation method, while prediction error of LSTM method is 261% less than RNS method. We classify producers into three types based on the prediction results and propose optimization measures aimed at the risks and challenges they faced. Field implementation indicates promising outcome with better reservoir support, lower GOR, lower WC, and stabler oil production. This study provides a novel direction for application of artificial intelligence in WAG flooding development and optimization.