Operators of offshore floating drilling units have limited time to decide on whether a drilling operation can continue as planned or if it needs to be postponed or aborted due to oncoming bad weather. With day-rates of several hundred thousand USD, small delays in the original schedule might amass to considerable costs. On the other hand, pushing the limits of the load capacity of the riser-stack and wellhead may compromise the integrity of the well itself, and such a failure is not an option.
Advanced simulation techniques may reduce uncertainty about how different weather scenarios influence the system’s integrity, and thus increase the acceptable weather window considerably. However, real-time simulations are often not feasible and the stochastic behavior of wave-loads make it difficult to simulate all relevant weather scenarios prior to the operation.
This paper outlines and demonstrates an approach which utilizes probabilistic machine learning techniques to effectively reduce uncertainty. More specifically we use Gaussian process regression to enable fast approximation of the relevant structural response from complex simulations. The probabilistic nature of the method adds the benefit of an estimated uncertainty in the prediction which can be utilized to optimize how the initial set of relevant simulation scenarios should be selected, and to predict real-time estimates of the utilization and its uncertainty when combined with current weather forecasts.
This enables operators to have an up-to-date forecast of the system’s utilization, as well as sufficient time to trigger additional scenario-specific simulation(s) to reduce the uncertainty of the current situation. As a result, it reduces unnecessary conservatism and gives clear decision support for critical situations.