| Geotechnical and Geo-environmental Graduate Award | Award | 2024-10-01 | Walid Ben Saleh |
| Coupled Flow-Geomechanics Surrogate Model with Flexible Boundary Conditions for Geological CO2 Storage Using Fourier Neural Operator Based Gated Recurrent Network University of Alberta | Publication | 2025-03-18 | Shuxin Qiao, Walid Ben Saleh, Zhang, B. |
| Geo-Resource Agent for Automated Reservoir and Mechanical Earth Model Characterization By Integrating Large-Language Models and Domain Specialized Toolbox AuthorsSubsurface resource development relies on complex workflows that integrate exploration, modeling, and reservoir management. These workflows are fragmented across software platforms and teams, and are increasingly challenged by the scale and heterogeneity of geological data. Large Language Models (LLMs) offer new opportunities for automation, but direct application is limited by token constraints and inefficiency in handling large numerical datasets. This paper presents the Geo-Resource Agent, a LangGraph-based framework that couples LLM reasoning with deterministic computational tools. The system follows a four-part design: an Agent Core for planning, a Tool Layer for domain-specific functions, a Retrieval Layer for schema- and unit-aware data access, and an Execution & Storage Layer for auditability. Numerical tasks such as log interpolation and decline curve analysis bypass the LLM, while the model focuses on coordination and schema alignment. Case studies demonstrate transparent reasoning, reduced manual scripting, and scalable automation for subsurface engineering workflows. University of Alberta | Publication | 2025-11-18 | Shuxin Qiao, Walid Ben Saleh, Ziming Xu, Vaniya Tariq, Zhang, B. |
| Geothermal reservoir modeling and calibration in a coupled thermo-hydro-mechanical perspective-a case study for DEEP Geothermal Project in Saskatchewan, CanadaGeothermal offers a low-emission, reliable and sustainable source of base-load energy, yet efficient reservoir management is challenging. Understanding the Thermo (T), Hydro (H) and Mechanical (M) interactions can help create sustainable strategies and ease energy extraction. In this study, a geological regional-scale THM model is created using PETREL, based on geological data from 11 wells close to the DEEP geothermal project. Geostatistical modelling was challenging as only 11 wells are available for such a large area. The depositional directions are used as secondary data for stratigraphic modeling. The permeability was populated using Machine Learning techniques based on 9 well logs as input data and calibrated with pulse-decay tests. A Mechanical Earth Model (MEM) was set up using a combination of empirical equations and calibrated using laboratory and field data. The minimum horizontal stress in the sandstone reservoir was found to be lower (~41 MPa) compared with caprock shale. Two open geothermal loop tests were history matched using INTERSECT which confirm fractures are induced for improved transmissibility in the sandstone reservoir without causing caprock failures.
University of Alberta | Publication | 2025-06-08 | Arqam Muqtadir, Zhang, B., Walid Ben Saleh, Chalaturnyk, R., Noga Vaisblat, "Andrew Wigston", "Ashley Drobot", Leo Groenewoud Groenewoud Leo, "Kristen Marcia" |
| Modeling with Confidence: Leveraging Conformal Prediction for Calibrated Machine Learning Based Mechanical and Petrophysical ModelsMachine learning (ML) is revolutionizing reservoir characterization practices by directly using well log and drilling data. However, deterministic predictions of ML models can be misleading and result in expensive mistakes. Hence, uncertainties in ML-predicted mechanical and/or petrophysical properties need to be well quantified. By leveraging a distribution-free and computationally efficient uncertainty quantification method called Conformal Prediction (CP), we can derive calibrated ML models and quantify the uncertainties in their predictions. Several ML models are developed for permeability and elastic modulus prediction in a geothermal site in south Saskatchewan. The Catboost model outperforms other models achieving an R2 of 0.91 and 0.92 for permeability and elastic modulus, respectively. A Conformal Prediction is then built on the selected ML models to complement the predictions with valid measures of prediction intervals with 95% coverage. For a test well, where two different lab-measured permeabilities exist, more than 90% of measured permeabilities fall within the 95% prediction interval. Triaxial geomechanical test results are also comfortably within the bounds of the 95% interval. This suggests that these models provide reliable predictions with limited uncertainties. This paper underscores the crucial role of uncertainty quantification of ML-based prediction models. The study demonstrates how quantifying uncertainty can enhance our confidence in ML-predicted reservoir properties for rigorous subsurface reservoir characterization. University of Alberta | Publication | 2025-06-08 | Walid Ben Saleh, Zhang, B. |
| Smart Coupled Flow-Geomechanical Upscaling Technique for Oil Sands University of Alberta | Publication | 2024-11-18 | Xiaoyan Ou, Shuxin Qiao, Zhang, B. |
| Smart and Dynamic Geological Carbon Storage Design Management Through Surrogate Model-Assisted Deep Reinforcement LearningOptimizing CO2 injection well placement and injection rate under geological and economic uncertainty is a key challenge for large-scale geological carbon storage. Unpredictable plume migration and pressure evolution in heterogeneous formations can lead to caprock failure, leakage, and uneconomic projects. Traditional optimization methods struggle with the high-dimensional design space and the large number of subsurface and economic scenarios. Deep reinforcement learning (DRL), in contrast, can adapt to such uncertainty through interaction with a simulated environment. In this work, a surrogate model-assisted DRL framework is proposed in which well placement is treated as a sequential decision problem. Each new injector is selected based on the evolving plume, pressure field, and remaining economic value. A policy is trained on surrogate-predicted CO2 saturation and pressure fields, together with sampled economic parameters, while excluding leakage-prone and infeasible regions via spatial masking. The objective is to maximize the discounted net present value (NPV) of the storage project by selecting 3 injector locations, subject to a pressure penalty that represents geomechanical safety limits and leakage penalty outside the leased boundary. University of Alberta | Activity | 2026-05-11 | Gamze Erdogan Erten, Zhang, B., "Jeff Boisvert" |
| Using Coupled Thermal-Hydraulic-Mechanical Simulation to Evaluate the Geothermal Power Extraction and CO2 Storage Capacity for Hybrid CO2-Water Geothermal SystemHybrid CO2-water geothermal systems offer a pathway to simultaneously extract geothermal energy and store CO2 in deep saline aquifers. However, their performance depends strongly on reservoir heterogeneity, CO2 breakthrough and plume development, and pressure evolution, all of which influence thermal output, flow sustainability, and geomechanical safety. Compared with conventional water-based geothermal systems, supercritical CO2 has lower viscosity, lower density, and stronger buoyancy forces, which may lead to higher production rates and less parasitic power required to pump the fluid to surface facility; however, the lower heat capacity of CO2 may also result in decreased thermal power production compared to water-based system.In this study, a hybrid CO2-water geothermal concept is evaluated using coupled thermal-hydraulic-mechanical (THM) simulations. Supercritical CO2 is injected to maintain reservoir pressure while enabling partial sequestration through structural, residual and solubility trapping. Production initially consists of brine and gradually transitions to a CO2-water mixture following CO2 breakthrough. The objective is to quantify the impact of CO2 breakthrough timing, fluid property contrasts, and reservoir heterogeneity on thermal power extraction, pressure management, and long-term system performance, and to compare the hybrid system against a conventional water-based geothermal configuration. Based on the THM simulation results, the hybrid CO?-water system produces comparable gross thermal power compared to water-based system after reasonable optimizations; however, CO2 benefits from strong buoyancy-driven circulation, which can significantly reduce parasitic pumping power. Over the 30-year operational period, 89 million tonnes of CO2 are stored in the saline aquifer. University of Alberta | Activity | 2026-05-11 | Arqam Muqtadir, Zhang, B., Chalaturnyk, R., Noga Vaisblat, "Andrew Wigston" |
| Corrosion control in carbon storage by injection of sodium formate solutionCarbon dioxide (CO2) leakage presents a significant risk to carbon storage in saline aquifers. Buoyant forces can cause CO2 in resident brine to migrate into overlying formations through faults, fractures, or existing wells. Wells are especially prone to leakage because CO2 produces an acidic environment, as recently observed in the Illinois Basin Decatur Project (IBDP). This study was motivated by the question of how to mitigate the risk of leakage caused by corrosion in well tubulars.
This research, for the first time, explores the use of sodium formate solution as a corrosion control method based on an IBDP geological model. To evaluate its effectiveness, reactive transport simulations of CO₂ and sodium formate solution injection were performed using the IBDP model. The simulation cases tested different injection rates (80 and 640 m³/d) and concentrations (5 and 15 wt.%), which resulted in varying amounts of sodium formate being injected. Both pre-flush and post-flush injection strategies were considered.
Results indicated that in the pre-flush scenarios, formate can effectively be dispersed by the subsequently injected CO2 (e.g., 150 to 350 meters from the well), while raising the pH above 4.5. However, to mitigate pH reduction near the injector, formate does not need to disperse widely; the scenario involving approximately 2.5 × 103 tonnes of sodium formate resulted in an average pH of 4.45 with a standard deviation of 0.27 along the injection well for 25 years, remaining safely above 4.0, which is the recommended pH range for corrosion control of 13 chrome steel pipe. University of Alberta | Publication | 2025-12-01 | Doguhan Barlas Sevindik, Oluwafemi Precious Oyenowo, "Ryosuke Okuno", Muhammad Farooq Zia, Zhang, B. |
| Deep reinforcement learning optimizing for geological CO2 storage considering geomechanical and leakage risksThe optimization of well placements and injection rates for maximizing CO2 storage remains a critical challenge for scaling up geological carbon storage (GCS) considering minimizing leakage outside of leased boundary and potential geomechanical risks. A deep reinforcement-learning (DRL) framework is developed in this paper, where a convolutional-neural-network (CNN) policy is trained to map surrogate-predicted CO2 saturation and pressure fields (states) to injection-well locations and dynamic injection rates (actions). Policy parameters are updated by Proximal Policy Optimization (PPO) with a multi-objective reward that increases with stored CO2 mass while imposing continuous penalties for pressure buildup and boundary leakage. Accurate surrogate models trained on coupled flow-geomechanics simulations significantly reduce computational cost and enable the agent and environment interactions required to train the DRL policy, supporting high-resolution evaluation across numerous geological realizations. The effectiveness of the proposed DRL framework is evaluated using a modified conditional geostatistical model based on the Aquistore CO2 storage site in Saskatchewan, Canada. Model-parameter uncertainties are handled using a domain randomization scheme, where multidimensional scaling and cluster analysis compress 1000 realizations into 20 clusters to streamline agent training. Performance is benchmarked against both fixed-rate reference configurations and the covariance-matrix adaptation evolution strategy (CMA-ES). Across all validation samples, the PPO-based policy consistently delivers superior well configurations and dynamic injection rates, achieving higher reward as injection volume while maintaining operational safety. On a representative validation sample, the PPO policy achieved 43.3 Mt of cumulative CO2 storage, significantly outperforming the 37.1 Mt achieved by the CMA-ES benchmark. University of Alberta | Publication | 2026-04-03 | Gamze Erdogan Erten, Zhang, B., Shuxin Qiao, Walid Ben Saleh, "Jeff Boisvert" |
| Development of a convolutional neural network based geomechanical upscaling technique for heterogeneous geological reservoir University of Alberta | Publication | 2024-06-01 | Zhiwei Ma, Xiaoyan Ou, Zhang, B. |
| Reducing geological uncertainty through coupled flow-geomechanics based surrogate models and rejection sampling of CO2 plume prediction University of Alberta | Publication | 2025-03-01 | Walid Ben Saleh, Zhang, B. |
