A Pre-Drilling Loss Risk Prediction Method Based on Seismic Data and Machine Learning

Abstract Losses of varying severity continue to pose a significant challenge in Middle Eastern drilling operations, primarily due to carbonate formations with complex loss channels and substantial heterogeneity. These losses are characterized by high levels of randomness and uncertainty, severely impacting drilling operations. To mitigate these risks, it is essential to predict loss risks before drilling and implement preventive measures in advance. Traditional loss risk analysis methods mainly rely on well logging and engineering data, which only allow for the prediction of losses during or after drilling. However, seismic data, available prior to drilling, can enable true pre-drilling loss prediction. The challenge with seismic data lies in the need for an inversion process to extract meaningful results from seismic waves, a process that can potentially generate an infinite number of solutions. This study addresses the issue by integrating seismic, wireline logging, and engineering data from both loss and non-loss wells, applying machine learning and intelligent algorithms to predict mud loss risks. The methodology begins by transforming seismic data into seismic reflection waves or rock physics using Python. Four seismic attributes- variance, time-frequency attenuation, sweet spot, and root mean square amplitude--are identified as highly correlated with loss occurrences. Next, acoustic and density logs are matched with seismic wavelets through de-convolution, aligning the time-domain seismic profile with the depth-domain logging data, and subsequently matching it with loss features recorded in existing loss wells. A pre-drilling loss prediction model is then developed using machine learning algorithms such as logistic regression, decision trees, and support vector machines, based on seismic, logging, and loss well engineering data. The model is continuously validated and tested with new data from loss wells until it achieves a high level of consistency with recorded loss events. Finally, a three-dimensional seismic loss probability volume (in SGY data format) is generated, which can be integrated into any platform that supports SGY volumes. This allows for flexible slicing and analysis of loss risk across different formations and regions. The proposed pre-drilling loss risk prediction method has been successfully applied, leading to optimized wells placements and the development of effective preemptive loss prevention strategies. This approach offers a powerful tool for reducing the risk of severe mud loss