Abstract
Accurate production forecast of multistage hydraulically fractured wells is crucial for the development of shale gas reservoir. While several data-driven production models have been proposed, few of them take into account the physical mechanism and production process.
In this paper, we present an innovative approach that combines domain knowledge with a deep learning algorithm to establish an accurate and interpretable production model. Integrated production influencing factors were extracted from four main aspects: geological reserve, fracture network shape, fracture conductivity and production control. Instead of treating the entire horizontal well as an average, features from individual stages were utilized as the model inputs. To account for the sequential gas aggregation from bottom to top stages, the gated recurrent unit (GRU) algorithm was applied, and an additional Mask layer was introduced to accommodate varying stage numbers among different wells.
The production model was trained with real data from 119 wells in Weiyuan shale gas field, China. The hyper-parameters were optimized using Bayesian optimization method, resulting in a robust performance with an average relative error (MRE) of 11.7%. This MRE is 77.4% lower than that of the traditional multilayer perceptron model. Furthermore, our model outperformed the simple GRU model by 37% in terms of MRE, demonstrating the significance of the Mask layer in avoiding data redundancy and improving information transmission efficiency. The results in this paper indicate the importance of considering input variables at the stage level and highlight the benefits of incorporating domain knowledge into production forecast.
Unlike conventional models that are solely data-driven, the approach proposed in this paper faithfully aligns with the actual production process of multistage hydraulically fractured wells, leading to significantly improved prediction performance. This study shows the potential of integrating domain knowledge and deep learning algorithm for production forecast in shale gas reservoirs.
