Production Prediction for Shale Oil Horizontal Wells Using Convolutional Neural Network–Bidirectional Long Short-Term Memory Neural Network Model

Shale oil represents an important unconventional resource in the global energy landscape, contributing significantly to the energy security. Predicting production from shale reservoirs, characterized by inherently low porosity and permeability coupled with complex postfracture networks, presents substantial challenges to conventional methods. To address these challenges, this study introduces a novel production prediction model based on the integrated application of convolutional neural networks (CNNs) and bidirectional long short-term memory (BiLSTM) networks, termed the CNN-BiLSTM model. This model synergistically leverages the spatial feature extraction capabilities of CNNs with the bidirectional temporal dependency capture of BiLSTM, enabling improved prediction accuracy for horizontal well production. Model validation involved a case study utilizing production data from horizontal wells in the Chang 7 Member of the Ordos Basin. The CNN-BiLSTM model's performance was systematically benchmarked against several machine learning algorithms: support vector regression (SVR), random forest (RF), extreme gradient boosting (XGBoost), long short-term memory (LSTM), and BiLSTM. Empirical results demonstrate the CNN-BiLSTM model's superior performance across all evaluation metrics, achieving a root-mean-squared error (RMSE) of 0.3103, a mean absolute error (MAE) of 0.1731, a mean absolute percentage error (MAPE) of 3.2333%, and a coefficient of determination (<i>R</i> ²) of 0.8856. This research offers an effective and robust approach for shale oil horizontal well production forecasting while also providing valuable theoretical and technical insights to inform development decisions for unconventional oil and gas resources.