Prediction of Gas Production Dynamic of Natural Gas Hydrate Reservoirs Based on Neural Network

As a clean and efficient resource with large reserves, natural gas hydrate has gained worldwide attention in recent years. Unfortunately, the productivity prediction of NGH reservoirs based on traditional numerical simulation is time-consuming and inefficient. It remains a great challenge to accurately and efficiently predict the productivity of NGH reservoir. In this study, a neural network model for predicting the gas production dynamic of hydrate reservoirs was established by learning the results of numerical simulations based on the geological parameters of hydrate reservoirs in Nankai Through of Japan. The accuracy of the neural network model was tested by comparing the actual production test of the hydrate reservoir in the Nankai Trough and it was employed to further predict the gas production dynamic of the hydrate reservoir. After testing, the alternative model for numerical simulation established using neural network has greatly improved the calculation speed and the accuracy rate exceeds 99.8%. In the first seven days, the average daily gas production of the hydrate reservoir in the Nankai Trough predicted by the neural network model is 21810.06 m³/d. The error between the predicted value and the actual value was less than 10%. Besides, the average daily gas production and cumulative gas production of hydrate reservoirs for one year were predicted to be 13800 m³ and 4.98 × 10⁶ m³ respectively by employing the established neural network model.