Integrated Petrophysical–Seismic Machine Learning Workflows for Dual-Purpose Reservoir Evaluation Using Kubernetes–OpenStack Infrastructure.

Abstract

The convergence of machine learning, cloud-native infrastructure, and geoscience workflows has created unprecedented opportunities for reservoir characterization at scale. This study presents an integrated framework that operationalizes Kubernetes–OpenStack container orchestration for dual-purpose reservoir evaluation, targeting both hydrocarbon productivity prediction and CO₂ storage suitability assessment. Building on validated infrastructure optimizations for GPU-intensive AI workloads in multi-tenant environments, this research demonstrates how containerized petrophysical and seismic machine learning pipelines can deliver measurable improvements in prediction accuracy, computational efficiency, and resource utilization. The proposed workflow integrates deep learning-based seismic attribute extraction, petrophysical property inversion, flow-unit classification, and storage capacity simulation within an autoscaling Kubernetes cluster deployed on OpenStack. Performance benchmarks reveal that GPU-accelerated training reduces model convergence time by 73% compared to CPU-only implementations, while container orchestration enables dynamic resource allocation that cuts infrastructure costs by 41% during peak workloads. The framework achieves 89.4% accuracy in porosity prediction and 86.7% in permeability estimation across heterogeneous carbonate reservoirs, while CO₂ storage capacity assessments demonstrate 92.1% agreement with conventional simulation methods at 18× faster execution speeds. By translating infrastructure-level efficiencies into domain-specific scientific outcomes, this work establishes a replicable methodology for deploying production-grade AI systems in computational geoscience, addressing the critical gap between cloud-native technology benchmarks and real-world reservoir engineering applications