Research and application of artificial intelligence-based prediction method for horizontal well-formation modelling

Horizontal well drilling has become an important method for oil companies to enhance single-well production in tight and unconventional oil and gas reservoirs. However, due to the complex spatial relationship between the wellbore trajectory of horizontal wells and the formation layers, traditional vertical well analysis methods cannot be effectively applied. Accurately describing the spatial combination relationship between the wellbore trajectory, the target layer, and the surrounding rock is a primary task in horizontal well logging interpretation. To address this issue, the mainstream approach is to construct an initial stratigraphic model based on a pilot well and then adjust the model segment by segment using forward modeling constraints from logging data. However, this process is time-consuming and requires numerous repetitive forward modeling calculations for different wells in the same area. Therefore, in the processing and interpretation of horizontal well logging data, developing a reasonable well-formation model is essential. This model enables an accurate description of the spatial relationship between the wellbore and the formation interfaces, including the distance between the wellbore position and formation interfaces and the angle between the wellbore axis and the formation normal direction. At the same time, logging data analysis methods based on machine learning and artificial intelligence (AI) technologies have been applied to various aspects of logging data interpretation by training intelligent models. With the support of AI technologies, it is expected to overcome the bottlenecks of traditional methods. To this end, the study proposed an automated horizontal well logging interpretation method based on multi-model integration and deep neural networks. First, a theoretical model was constructed incorporating different wellbore trajectories and formation combination relationships, and a logging response sample library was generated. Then, machine learning models such as eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost) were integrated, and their prediction results were further fused using a multi-layer perceptron (MLP). Finally, intelligent automatic recognition of the geometric relationship between the well trajectory and the surrounding rock was carried out using actual logging data. Case analysis showed that this method accurately captured the complex logging response characteristics of horizontal wells while significantly improving interpretation speed and accuracy. The proposed method meets the demand for rapid analysis of multiple wells in similar geological environments and provides an efficient, intelligent approach to horizontal well logging interpretation.