Abstract
Many of the 12,000 offshore oil and gas platforms are approaching the end of their lifecycle and the decommissioning decision for these structures is crucial. Multi-criteria Decision Analysis (MCDA) is frequently employed to determine decommissioning options. However, MCDA presents significant inefficiencies including subjectivity, prolonged consultation period and substantial resource requirement. This study proposes an explainable AI (XAI) approach to address these inefficiencies in predicting a suitable decommissioning option using a historical dataset from the United Kingdom. After splitting the dataset into the SURF subset (umbilicals, pipelines and cables) and the Non-SURF subset consisting of all other asset types, five machine learning models-Random Forest, Decision Tree, Neural Networks, Support Vector Machine (SVM) and K-Nearest Neighbours (kNN) were trained on each subset. The Random Forest model demonstrated superior performance for the Non-SURF subset, achieving a weighted average F1-score of 84%, while the Decision Tree model performed best with the SURF subset with a score of 79%. To further explore the explanation in support of each de-commissioning option recommended by the models, Local Interpretable Model-agnostic Explanation (LIME) has been applied. LIME identifies " Position " , " Metal " , " diameter " and " Residues " as important contributing predictors. In both the SURF and Non-SURF cases, the LIME explainer identified the position of the asset as a key factor in choosing the decom-missioning option. Other factors included in the decision-support are the presence of metal for most of the Non-SURF structures and the presence or absence of residues for the SURF structures.