Geophysical Well Logging in the AI Era: Practices and Prospects

Artificial intelligence (AI) technology has made significant progress, attracting considerable attention, and its applications are expanding across all industries. This article introduces the current state of AI applications in the oil and gas sector, analyzes their development trends, and presents preliminary progress in the exploration and research of AI application scenarios within the field of well logging. Constructing an effective intelligent well logging application scenario hinges on clearly defining the scenario's purpose, problem statement, business logic, input/output requirements, as well as the source and justification of the dataset and labels. The effectiveness of discriminative machine learning is fundamentally determined by the dataset and the labeling system. Well logging data possesses clear physical significance but is costly to acquire. It is influenced by factors such as the wellbore, invaded zone, surrounding rock, and the instrument's varying depths of investigation, vertical resolutions, and circumferential response characteristics. Meanwhile, core sample analysis and perforation/production testing are constrained by scale heterogeneity. These factors make effective point-by-point processing, interpretation, and labeling challenging. Consequently, generating well logging data via forward modeling becomes both necessary and feasible. The project team explored nine types of application scenarios: well log quality control, core depth matching, depth alignment, reservoir parameters prediction, cement bond evaluation, near-wellbore fracture identification, cased-hole logging evaluation, nuclear magnetic resonance (NMR) logging, and well log data generation. Scenario construction, model training, and testing have been completed. Among these, reservoir parameters prediction and cement bond evaluation have been deployed at scale and have yielded practical results.