A CWGAN-GP Based Efficient Proxy Model for Predicting Pressure Profiles in Oil Reservoir

Accurate prediction of reservoir pressure can timely obtain corresponding adjustment countermeasures, which is of great significance for scientifically and rationally formulating mining plans and improving recovery rates. Usually, the production dynamics prediction adopts numerical simulation methods, but the complexity of reservoir geological conditions will greatly increase the difficulty and time required to achieve a complete simulation. The rapid development of technologies such as big data and artificial intelligence has provided a new perspective for solving this problem. Based on the theory of CGAN, the scope of application of the model is improved, and six conditional parameters including time, matrix permeability, and fracture permeability are added to achieve real-time dynamic prediction of reservoir pressure under any conditional parameters. The increase in conditional parameters improves the generalization of the model, but also increases the difficulty of convergence of the model. Therefore, the loss function is optimized by combining the Wasserstein distance and gradient penalty theory, forming a reservoir pressure prediction proxy model that is extremely easy to converge and has strong generalization. This model considers the influence of multi-scale reservoir uncertainty and well control changes, and successfully predicts the pressure at any time under fracturing horizontal well production, with an accuracy of more than 98%, and an efficiency that is 2-3 data levels higher than numerical simulation. It can be used to solve computationally intensive tasks such as history matching and production optimization, and also provides new ideas for the construction of big data models for dynamic prediction of unconventional reservoirs.