Metahueristically Optimized Graph Attention – Transformer Model for Carbon Emission Modelling in Oil and Gas Operation
Year: 2026
Type: conference-paper
Abstract: The oil and gas industry needs accurate predictions of carbon dioxide$(\mathbf{C O}_{\mathbf{2}})$emissions in order to maintain sustainability. This study present a hybrid framework composed of the American Zebra Optimization Algorithm (AZOA) in addition to the Graph Attention Network- Transformer model (GAT-Transformer) for predicting carbon emissions. The framework effectively combines the benefits of both algorithms through several data-related processes that help increase the accuracy of predictions. During the first stage of the framework, explore the dataset by cleaning the data and identifying patterns, this process helps to improve the quality of data prior to conducting any analyses. Next, it performs feature engineering to reduce the number of dimensions in dataset and extract the most important features associated with emissions, thus eliminating redundant information and retaining only the relevant information related to emissions. Once the dataset has been pre-processed, it is split into training and testing subsets before classifying it. This ensures that the learning process is not biased and the model's performance accurately evaluated. The GAT-Transformer model captures both temporal dependence and complex spatial relations among emission data, while AZOA helps achieve optimal hyperparameters tuning and, therefore, increases the rate and accuracy of the algorithm's convergence. Based on the results of our experiments, which were conducted in Python, we achieved a MAE of 0.074, a RMSE of 0.1442, and a MSE of 0.0208; all of these results validate the robustness and effectiveness of the proposed framework for monitoring and supporting decision-making related to carbon emissions.
Language: English
Field: Energy
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204848218