Oil production prediction algorithm based on TRIZ conflict resolution and hybrid neural network model

• Theory of Inventive Problem Solving guides the model design. • A hybrid neural network predicts oil production more accurately. • The model combines spatial, temporal, and global data features. • It was validated on real field data from multiple oil wells. • This approach balances prediction accuracy with computational cost. Oil production forecasting serves as a critical basis for optimizing oilfield development decisions. Addressing limitations in existing methods—such as limited accuracy in time-series forecasting and poor interpretability—this paper proposes a novel oil production forecasting approach integrating the Theory of Inventive Problem Solving (TRIZ) with deep learning. First, the TRIZ contradiction matrix system is employed to analyze typical engineering contradictions in oil production forecasting, including “accuracy improvement versus increased model complexity” and “reliability versus time cost.” Based on this analysis, inventive principles such as separation, combination, and partial action are derived, providing theoretical foundations for model construction. Building upon this, a combined model incorporating Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Units (BiGRU), and Transformers. CNN extracts local spatial features from production data, BiGRU models bidirectional temporal dependencies, while a lightweight Transformer—trimmed to remove redundant structures—integrates global information and outputs predictions. Using historical data from a specific oilfield as samples, experiments demonstrate that the proposed model outperforms comparison models across three metrics: Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). Specifically, RMSE is 9.78, MAE is 5.45, and R² reaches 0.91. Furthermore, this model significantly reduces computational overhead while maintaining prediction accuracy, validating the effectiveness of TRIZ-guided structural improvements. This research not only provides a high-precision oil production forecasting model but also establishes a TRIZ-based, systematically reusable deep learning modeling workflow. It offers a structurally clear and interpretable technical approach for intelligent oilfield forecasting.