Research on Yield Prediction Model Driven by Mechanism and Data Fusion

Area of Science:

• Petroleum Engineering
• Artificial Intelligence
• Data Science

Background:

• Traditional production forecasting methods lack accuracy due to limited data sources and poor integration of physical principles.
• Existing models often fail to capture complex dynamics in oilfield production time series.

Purpose of the Study:

• To develop a robust production forecasting model by fusing mechanistic simulations with data-driven techniques.
• To improve the accuracy and reliability of oil production predictions in the petroleum industry.

Main Methods:

• A three-phase-separator mechanistic model was developed to generate physics-informed data.
• A Global-Local Branch Prediction Model was designed to capture both long-term trends and local features.
• Mechanistic model outputs were integrated as constraints within the data-driven prediction framework.

Main Results:

• The proposed mechanism-data fusion model demonstrated superior performance compared to state-of-the-art methods like Autoformer and DLinear.
• Significant reductions in Mean Squared Error (MSE), Mean Absolute Error (MAE), and Root Squared Error (RSE) were achieved.
• The model reduced MSE by 0.0100 and MAE by 0.0501 compared to Autoformer, and improved MSE by 0.0080 and MAE by 0.0093 over DLinear.

Conclusions:

• Integrating mechanistic constraints into data-driven models substantially enhances production forecasting accuracy.
• The proposed Global-Local Branch Prediction Model offers a technologically superior and robust solution for petroleum engineering applications.