Prediction model of lost circulation based on drilling parameters with PSO-BP neural network

Abstract Accurate and real-time prediction of lost circulation is essential for ensuring drilling safety and operational efficiency. Existing data-driven approaches, however, often lack effective integration of domain expertise, which limits their reliability in field applications. To bridge this gap, this study develops an intelligent hybrid model (PSO-BP) that systematically incorporates field-based insights into a machine learning framework. The key contributions are twofold: first, through expert-guided analysis of comprehensive mud logging data, characteristic parameters, including total pit volume, standpipe pressure, flow-in/out difference, and top drive load, are identified as effective indicators of lost circulation; second, a novel prediction model is established by employing Particle Swarm Optimization (PSO) to optimize the initial weights and thresholds of a Backpropagation (BP) neural network, thereby enhancing its convergence speed and predictive stability. Compared with standard BP, Beetle Antennae Search (BAS)-BP, and Genetic Algorithm (GA)-BP models, the proposed PSO-BP model demonstrates superior accuracy in forecasting lost circulation incidents. Validation under actual drilling conditions confirms its practical effectiveness. This research provides a robust and interpretable tool for early loss detection, contributing significantly to risk mitigation, cost reduction, and overall drilling efficiency.