<![CDATA[Flowing Bottomhole Pressure (FBHP) is an essential factor for oil well performance evaluation, but conventional measurement methods can be costly and lack real-time capability. This study presents a machine learning approach to estimate FBHP using simulated data from established vertical flow correlations. The proposed framework includes four main steps: collecting input parameters, simulating pressure drops calculation, developing an artificial neural network (ANN) model, and designing the FBHP calculation algorithm. The ANN was developed using key input variables, including inlet pressure, system temperature, tubing size, inclination, segment length, gas-oil ratio (GOR), water cut, oil API gravity, gas gravity, fluid rate, and vertical flow correlation type. A dataset of 790,409 points from several multiphase flow simulations was used, covering various well conditions for naturally flowing oil wells without artificial lift. The optimal ANN architecture featured six hidden layers and was trained with transformed, encoded, and normalized inputs, achieving a testing mean absolute error (MAE) of 7.8259 psia and R² of 0.9993. Segment-level predictions are then conducted iteratively to estimate FBHP for the whole well trajectory. Compared to earlier studies, the novelty of this work lies in its large and diverse set of well-flowing conditions, combined with comprehensive tubing geometry using segmentation. This approach enables the modelling of a wider range of flow scenarios and complex well trajectories.]]>
