Enhanced coalbed methane well production prediction framework utilizing the CNN-BL-MHA approach

Area of Science:

• Petroleum Engineering
• Data Science
• Artificial Intelligence

Background:

• Coalbed methane (CBM) extraction faces challenges due to evolving geology and mechanization, leading to non-linear production data.
• Traditional single deep-learning models struggle with CBM production prediction due to issues like overfitting and gradient instability.

Purpose of the Study:

• To develop an advanced model for accurate CBM gas production prediction.
• To overcome the limitations of single deep-learning models in predicting complex CBM production trends.

Main Methods:

• A novel CNN-BL-MHA model was developed, integrating Convolutional Neural Networks (CNN) for feature extraction with Bidirectional Long Short-Term Memory (Bi-LSTM) and a Multi-Head Attention (MHA) mechanism.
• The model utilized production data from Wells W1 and W2 for training and prediction.
• Performance was benchmarked against single models including ARIMA, LSTM, MLP, and GRU.

Main Results:

• The CNN-BL-MHA model demonstrated superior accuracy and stability in predicting CBM gas production.
• Prediction accuracy improved by up to 35% compared to single deep learning models.
• The model's predictions closely matched actual yield data with reduced error.

Conclusions:

• The CNN-BL-MHA model offers a significant advancement in CBM gas production forecasting.
• Its enhanced accuracy and stability make it a reliable tool for the CBM industry.
• This approach effectively addresses the challenges posed by non-linear production data in CBM wells.