Identification of Interwell Connectivity and Prediction of Oil Production Index in Fractured-Cave Oil Reservoirs

Abstract Fractured-cave carbonate reservoirs are typically characterized by strong heterogeneity. Accurately identifying interwell connectivity and formulating appropriate development strategies are crucial for the effective exploitation of these reservoirs. This study establishes a set of comprehensive methods to address these challenges. Static data such as seismic, drilling, logging and mud logging data, as well as geological modeling results, are used to determine the types and distribution of reservoir bodies around wells. Based on dynamic data including production data, pressure, and tracers, pressure transient analysis and dynamic parameter indicator curves are applied to evaluate the interwell connectivity and the types of connected reservoir bodies. Finally, a comprehensive analysis integrating static and dynamic data is carried out to determine the interwell connectivity. Based on the analyzed interwell connectivity relationships, the fracture-cave combination relationships are identified, and corresponding models are established. The impacts of factors such as pressure differential, water cut, and fracture permeability on the oil production index are simulated.Taking the TH oilfield in China as an example, we utilize static data to identify the reservoir types and the distribution of reservoirs around wells, and dynamic data is applied to evaluate the interwell connectivity. Subsequently, a multiple linear regression model is adopted to quantitatively evaluate the interwell connectivity of typical connected well groups. Based on the analyzed interwell connectivity relationships, a typical cave - fracture - cave model is identified. Numerical simulation methods are used to study the effects of pressure differential, water cut, and fracture conductivity on the oil production index. Subsequently, eight machine learning algorithms are applied to construct a prediction model for the oil production index. Four evaluation indicators are used to optimize the prediction models, and finally, Light Gradient Boosting Machine (LightGBM) is selected as the prediction model for the oil production index. The prediction results of this model are compared with the actual oilfield data, showing that the error is less than 5%. These insights provide technical support for analyzing the utilization status of developed units and designing new well trajectories to connect more reserves and ensure the profitable development of the oilfield.