Real-Time Prediction of Bottomhole Equivalent Circulating Density (ECD) Based on Machine Learning Algorithms in Offshore Deepwater Drilling

Maintaining bottomhole pressure within the safe range is crucial for deepwater drilling operations. The equivalent circulating density (ECD) of the drilling fluid at the bottom hole serves as an indicative measure of wellbore pressure. However, accurately predicting the ECD of drilling fluid is challenging due to the complex variations in wellbore temperature and significant changes in drilling fluid properties encountered in deepwater drilling. Currently, obtaining ECD primarily depends on model calculations and downhole detection equipment. However, these methods are not widely applicable due to their cumbersome calculations and high costs. The maturation of machine learning theory offers a new avenue for accurately predicting ECD. This study employed the Generalized Regression Neural Network (GRNN) to train a model based on the time series data of drilling characteristic parameters. A network search, coupled with an intelligent optimization algorithm, was employed to optimize the model's hyperparameters. Another dataset was chosen to assess the model's generalization capability. The reliability and accuracy of the model were verified by comparing it with measured and simulated data. Upon verification, the prediction model's hyperparameters (smoothness factor δ) achieve the optimal performance at 1.0657. Compared with the actual values, the final prediction result of the model has a small error, with a root mean square error (RMSE) of 0.3615 and an average absolute percentage error (MAPE) of 2.43%.The findings of this research provide valuable references and guidance for optimizing hydraulic parameters and controlling wellbore pressure in deepwater drilling.