Fine-Scale Identification of Deep Lithology and Potential Favorable Intervals in WWY1 Well, Wuwei Depression: Implications from Support Vector Machine Analysis of Multiparameter Logging Responses

Abnormally overpressured helium-rich natural gas occurs in dolomite reservoirs beneath gypsum-bearing strata of the Middle Triassic Zhouchongcun Formation in WWY1 Well, Wuwei Depression. Resolving the vertical arrangement of dolomite, gypsum, and shale is therefore critical for evaluating reservoir–seal coupling in this structurally complex setting. Here, we develop a support vector machine (SVM) workflow using acoustic transit time (AC), bulk density (DEN), compensated neutron log (CNL), gamma ray (GR), and spontaneous potential (SP) to refine lithology classification and support lithology-based screening of potential favorable intervals. The reference lithology column integrates mud-logging descriptions, conventional log responses, and stratigraphic information. An RBF-SVM was optimized by grid-search cross-validation, and formation-specific constraints were introduced to suppress lithologies incompatible with the local stratigraphic association. The unconstrained and formation-constrained models yielded matching rates of 89.16% and 90.23%, respectively. Although the numerical increase is modest, the constrained model reduces cross-formation confusion, improves boundary continuity, and produces a more geologically coherent representation of the gypsum–dolomite–shale succession. The model also delineates several thin sublayers supported by synchronous multiparameter anomalies. Independently, epsilon-SVR reconstruction of U, Th, and K logs achieved R2 values of 0.979–0.998 and restored continuous element-specific radioactivity information across the interval lacking spectral gamma-ray measurements. Together, these results establish a high-resolution lithological framework for characterizing reservoir–seal architecture and screening dolomite-dominant intervals beneath effective evaporite seals as potential exploration targets. The workflow is designed for single-well geological refinement rather than universal model benchmarking; final evaluation of reservoir quality requires integration with porosity, fracture, core, image-log, pressure, and production-test data.