【摘要】 基于数值模拟求解的传统水合物藏产能预测方法耗时长、效率低，准确、高效地预测天然气水合物藏产能是目前面临的难点问题。基于实际海域水合物藏的地质数据建立大量数值模拟样本，利用神经网络模型对数值模拟结果进行学习，建立水合物藏产能预测的神经网络模型。同时，预测神狐水合物藏和日本南海水合物藏开采2 a的产能，并对降压开采的压力进行推荐。结果表明：神经网络模型预测的准确率超过97%;预测得到神狐水合物藏2 a的日均产气量为2 839 m~3,最佳开采压力为3 MPa;日本南海水合物藏2 a的日均产气量为21 523 m~3,综合考虑产气量和气水比，最佳开采压力为4 MPa;69%的水合物藏适合的开采压力为3 MPa;但当水合物藏的水合物饱和度大于65%、地层绝对渗透率高于0.1μm~2及原始地层压力高于20 MPa时，建议开采压力选择5 MPa。更多还原

【Abstract】 In order to develop a fast and efficient numerical simulation technique for productivity prediction of natural gas hydrate reservoirs, a neural network model was established by learning the results of conventional reservoir simulations, in which the learning samples of the new model were established based on the geological data of actual hydrate reservoirs. The model was applied for a two-year period of production prediction of the hydrate reservoirs in the Shenhu and the Nankai Trough regions in terms of depressurization. The simulation results indicate that the prediction accuracy of the neural network model exceeds 97% in comparison with the conventional reservoir simulators. The predicted average gas productions rate for the Shenhu hydrate reservoir is 2 839 m~3/d, with an optimal production pressure of 3 MPa. For the hydrate reservoir in the Nankai Trough, the predicted two-year average gas productions rate is 21 523 m~3/d, and the optimal production pressure is recommended as 4 MPa. The results of productivity prediction for various hydrate reservoirs indicate that production pressure of 3 MPa is suitable for nearly 69% of the hydrate reservoirs studied. However, there commended production pressure can be better of 5 MPa for reservoirs with hydrate saturation greater than 65%, the absolute permeability higher than 0.1 μm~2 and the initial reservoir pressure higher than 20 MPa.更多还原