Abstract
<jats:p>The article presents the development of a hybrid CNN-LSTM neural network model for intelligent optimization and early prediction of complications in oil and gas well drilling processes. The relevance of the research is driven by the increasing complexity of geological and technical conditions, the high cost of drilling operations, and the necessity to minimize risks associated with emergency situations such as drill string sticking, mud losses, and excessive mechanical loads. The drilling process is considered a multidimensional nonlinear dynamic system that generates large volumes of real-time telemetry data. Traditional analytical and physics-based models often fail to provide sufficient prediction accuracy due to the complex interaction between technological parameters and geological factors. The proposed model integrates Convolutional Neural Networks (CNN) for automatic extraction of local spatiotemporal features with Long Short-Term Memory (LSTM) networks capable of capturing long-term dependencies in drilling time-series data. This hybrid approach enables simultaneous telemetry noise reduction, identification of characteristic patterns in torque, pressure, and weight-on-bit variations, and forecasting of the Rate of Penetration (ROP), as well as the probability of operational complications. The paper provides a mathematical formulation of a multi-objective optimization problem that accounts for productivity, energy efficiency, and operational risk constraints. A combined loss function is introduced to incorporate both prediction accuracy and proximity to critical technological thresholds. Data preprocessing techniques are described, including outlier detection and correction, digital signal filtering, normalization, and sliding-window segmentation for training dataset preparation. Special attention is given to model training under conditions of class imbalance and stochastic disturbances. The results indicate that the CNN-LSTM architecture significantly enhances the predictive performance of the intelligent decision support system and enables early detection of potentially hazardous drilling trends. The findings confirm the feasibility of applying deep learning methods to the development of intelligent drilling management systems and establish a foundation for further implementation of digital twins and adaptive optimization technologies in the oil and gas industry.</jats:p>