Early detection of overflow based on genetic algorithm for capturing multiple feature changes in managed pressure drilling

Drilling overflows can result in significant losses of money and personnel. So early detection of overflows is of great practical importance. In this paper, six managed pressure drilling features related to overflow are selected, outlet and inlet flow difference, standpipe pressure, methane, ethane, pump flush and hook height. Then the variance, slope and mean of each feature within a statistical time window are calculated. Their thresholds are optimised to detect the overflow point according to the change of statistics by an improved multi-objective genetic algorithm. Logistic chaotic mapping is used to initial the genetic algorithm, and the Lévy flight is employed to improve the mutation operator. Experiments show that the new algorithm achieves an average overflow recall rate of 93.7%. The method is able to provide early warning for drilling engineers, thus further safeguarding wellbore safety. [Received: April 7, 2024; Accepted: June 6, 2024]