Research & Technology

Physics-informed Neural ODE for continuous soil moisture dynamics, paired with the GE-Water distributed hydrological engine. Validated on 21 storm events in Seocho-gu with R² = 0.983.

Dynamic Green-Ampt IMD coupling that overcomes static AMC limitations. Built on a 2,497-conduit GIS trunk sewer network for Seocho/Gangnam-gu — Mixed LID at 40% coverage yields 39.6% annual runoff and 33.8% peak-flow reduction.

XGBoost surrogate (R²=0.79) + NSGA-II multi-objective optimization shows that 30–50% permeable-pavement coverage is the cost-effective sweet spot, with SHAP identifying high-impact sub-districts.

GATv2 trained on 1,500 OpenFOAM scenarios. Seocho → Gangnam transfer learning achieves zero-shot R²=0.845, few-shot R²=0.887. Bayesian optimization identifies dense street-tree planting (5–8 m, LAI 4–5) as the top cooling strategy (up to 1.98 °C per block).

Integrated ABAQUS dynamic FE + Kaimal-spectrum wind + rainflow + Miner's rule + Weibull. Quantifies fatigue-life differences across five Korean street-tree species (Korean Ministry of Environment R&D 2022003570004).

Relaxes the classical concentric-cavity assumption with 10,000 Monte Carlo eccentric geometries. A first-order Sobol decomposition attributes 99.6% of variance to normalized eccentricity — a statistical basis for prioritizing sonic-tomography surveys.