Automatic Characterization of Effective Fracture Geometry and Production Profiles in Horizontal Wells Using High-Resolution Distributed Strain Sensing

Summary Accurate characterization of fracture geometry and production profiles is critical for optimizing completion designs in unconventional reservoirs. Rayleigh frequency shift (RFS) distributed strain sensing (DSS) is a powerful diagnostic tool for mapping fracture geometries and identifying production profiles along horizontal wells. Although strain data from the B4H well in the Hydraulic Fracturing Test Site 2 (HFTS-2) project were collected, comprehensive cluster-level analysis remains unperformed. In this study, we introduce a multiproxy-based workflow to automate strain-data history matching for two completion designs (6 vs. 10 clusters per stage) along the entire B4H wellbore. The proposed workflow reduces uncertainty, quantifies fracture geometries, and identifies production profiles, enabling robust completion optimization. We analyze RFS-DSS strain data acquired during the shut-in period of B4H and evaluate two completion designs applied across 26 stages and 172 clusters: 6 clusters per stage and 10 clusters per stage. Strain change attributes are assessed at the levels of completion designs, stages, and clusters. A detailed comparative analysis of strain attributes between the two designs is conducted. A machine learning–based Markov Chain Monte Carlo (MCMC) workflow is developed for automated history matching of strain responses along the wellbore. Following history matching, effective fracture properties are estimated with associated uncertainties. These calibrated fracture geometries are then used to evaluate production profiles at the cluster level throughout the well for both completion designs. The fracture properties and production profiles are compared between the two completion designs to evaluate the efficiency of the completion design. Among the 172 designed clusters in the B4H, significant strain changes are observed in 159 clusters. Distinct differences exist between the two completion designs: The 10-cluster stages generally exhibit higher peak strain values (median ~32 με) compared with the 6-cluster stages (median ~22 με), while the 6-cluster stages present wider strain widths (median ~15 ft vs. ~11 ft). A machine learning–based history-matching workflow is applied to reproduce the observed strain profiles and calibrate fracture geometries at the cluster level. On average, the 6-cluster stages exhibit longer fracture half-lengths (median ~200 ft vs. ~160 ft), larger stimulated reservoir volume (SRV) per cluster (median ~2.5 ft vs. ~1.0 ft), and slightly higher fracture heights (median ~180 ft vs. ~150 ft) compared with the 10-cluster stages. While the median fracture conductivity remains similar between the two, greater heterogeneity is observed in the 10-cluster design. Results show that the 6-cluster completion design generates more effective fracture geometries with greater spatial extent. Based on the calibrated fracture geometries, the 6-cluster stages show more uniform production profiles and slightly higher median normalized oil rate per cluster compared with the 10-cluster stages in this data set. This study pioneers the systematic analysis of strain responses measured along an entire horizontal producing well during the shut-in period in unconventional reservoirs, leveraging machine learning–based techniques. The results deepen our understanding of field strain measurements, enhance the usage of field data for production profiling, and enable effective fracture geometry characterization at the cluster level. Furthermore, the findings offer valuable insights into the impact of different completion designs on fracture geometry and production distribution, contributing to the optimization of completion strategies and the accurate estimation of production performance using fiber-optic strain measurements in unconventional reservoirs.