Machine Learning-Based HT-ATES Simulation | #HTATES #ThermalEnergyStorage #AquiferStorage #EnergyModeling #SustainableHeating #ClimateTech #DataDrivenModeling
1. Introduction
-
Climate Challenge & Heating Sector: Heating accounts for ~40% of global energy demand and is a major contributor to CO₂ emissions.
-
Role of HT-ATES: High-Temperature Aquifer Thermal Energy Storage (HT-ATES) can reduce fossil fuel use by storing excess heat for use during peak demand.
-
Need for Efficient Modeling: Traditional numerical models (FEM, FDM, FVM) are accurate but computationally intensive (1–10 hours per run), making large-scale integration impractical.
2. Modeling Approaches for HT-ATES
2.1. Numerical Modeling
-
Finite difference, finite volume, and finite element methods simulate fluid and heat flow accurately.
-
Require complex setups and significant runtime.
-
Unsuitable for real-time or large-scale energy system planning.
2.2. Analytical Modeling
-
Only one known analytical method exists.
-
Fast but inaccurate across wide parameter ranges (e.g., changes in screen length, injection rate).
-
Limited use in energy system simulations.
3. Data-Driven Model (DDM) for HT-ATES
3.1. Overview of the DDM
-
Predicts temperature profiles based on 7 key aquifer parameters.
-
Combines Machine Learning (ML) with a Nearest Neighbor Search to ensure efficiency and accuracy.
-
Replaces computational physics simulations with a data-based approximation.
3.2. Role of Machine Learning
-
Extreme Gradient Boosting (XGBoost) is used to predict Recovery Efficiency (RE).
-
Trained on 3501 simulation data points covering diverse parameter ranges.
-
Validated using RMSE; ML outperforms linear interpolation significantly.
3.3. Nearest Neighbor Search for Profile Matching
-
Uses RE, injected temperature, and ambient temperature to find the closest match from the dataset.
-
Ensures physical realism by basing predictions on simulated full-physics results.
3.4. Profile Adjustment
-
Temperature profiles are adjusted to match new injected temperatures.
-
Converted to temperature-over-volume format to generalize across operation profiles.
4. Model Evaluation
4.1. Accuracy Testing
-
ML predicted RE with RMSE of 1.45 percentage points vs. 16 for linear interpolation.
-
DDM overall produces low error when compared with original numerical models.
4.2. Parameter Sensitivity
-
Tested different nearest-neighbor distance metrics (using subsets of parameters).
-
Best performance achieved using RE, injection temperature, and ambient temperature.
4.3. Limitations Identified
-
Accuracy drops when using operation profiles vastly different from the base (e.g., mixed extraction/injection).
-
Some heat loss behavior tied to time is not captured in volume-based profiles.
5. Application Potential and Use Cases
-
Energy System Integration: Enables HT-ATES to be included in broader energy planning tools.
-
Scenario Analysis & Optimization: Fast simulation allows thousands of runs for uncertainty and sensitivity analysis.
-
Sustainable Heating Systems: Helps design systems combining solar/geothermal with seasonal heat storage.
6. Conclusion
-
The proposed DDM enables fast, accurate temperature profile prediction for HT-ATES systems.
-
It fills a critical gap between physics-based modeling and energy system simulation.
-
Open-source implementation makes it ready for adoption in research and planning tools.
Comments
Post a Comment