基于物理经验模型约束的机器学习方法在页岩油产量预测中的应用

页岩油气产量预测是确定其开发经济性的重要手段，目前的产量预测研究很少能在物理模型与数据挖掘方法之间达到统一.针对页岩油气的产量分析，本研究深入结合误差反向传递（BP）神经网络和长短期记忆（LSTM）神经网络的数学方法优势，综合考虑工程经验模型的约束，改善了模型预测精度，经过实例数据训练后可较好地预测油田产量，并研究了页岩储层深度、总有机碳含量（TOC）、脆性度等油田参数对产量预测的影响规律.这项工作可以为页岩油气规模化开发提供可靠的产量预测和经济评价.

Machine Learning With Physical Empirical Model Constraints for Prediction of Shale Oil Production

Prediction of oil and gas production is an important way to determine its development economy. However, at present the production prediction is still hard to achieve consistency between the physics-based method and the data-based method. For shale oil and gas production analysis, in-depth combination of mathematical advantages brought by BP neural networks and LSTM neural networks, and comprehensive consideration of physics-based models, lead to good improvement in the prediction accuracy of the model. After training with practical testing data, the prediction of oilfield production can be significantly improved. Afterwards, the effects of the reservoir depth, the TOC and the brittleness, etc. on production prediction were studied. In conclusion, the work provides reliable production prediction and economic evaluation for large-scale development of shale oil and gas.