The traditional method of distributary channel system (DCS) reservoir modeling has weak adaptability to dynamic data and is difficult to effectively characterize heterogeneity and complex well network configuration issues. In order to improve the accuracy and efficiency of the DCS reservoir modeling, this study proposes an improved Monte Carlo Tree Search (MCTS) algorithm based on the reinforcement learning. This method enhances the comprehensive adaptability to geological constraints and dynamic responses in the modeling process by dynamically constructing a search tree, synchronously balancing global and local optimization, and introducing optimization mechanisms such as the cycling removal. Case studies show that the quality of this reservoir model established by using this method is equivalent to those of models manually constructed by geological experts, and is superior to traditional modeling methods in terms of prediction accuracy and calculation efficiency, providing effective technical support for making detailed reservoir modeling and development decisions of the oil and gas fields.
ZHANG et al. (Thu,) studied this question.