An Efficient Approach for Automatic Complex Fractured Networks Parameter Inversion Based on Surrogate Model and Deep Reinforcement Learning

Abstract

AbstractComplex fracture networks are commonly generated in subsurface reservoirs, and the first and primary task is to understand their transport characteristics and properties. Like pumping test interpretation, the well‐testing interpretation is an effective means for transport characteristics analysis and parameter estimations, this work attempts to propose a parameter inversion approach for complex fracture networks based on well‐testing interpretation. Given that non‐unique solutions and computational inefficiencies are obstacles to practical interpretation, especially when complex fractures are existing, a new deep reinforcement learning (DRL) based approach is proposed for automatic curve matching on complex fractured networks well‐testing interpretation. Based on twin delayed deep deterministic policy gradient (TD3) algorithm, the proposed DRL approach is successfully applied to automatic matching of complex fractured networks' well test curves. In addition, to improve the training efficiency, a surrogate model of the well test model based on bidirectional GRU (Bi‐GRU) neural network was established. After episodic training, the agent finally converged to an optimal curve matching policy through interaction with the surrogate model. The results show that the average relative error of the curve's parameter interpretation is 4.34%. Additionally, the results from the case studies show that the proposed DRL approach has a high calculation speed.