Explainable AI-driven prediction and interpretation of aerodynamic interference effect in complex high-rise building clusters

Abstract

The rapid urban densification intensifies aerodynamic interference among high-rise buildings, which complicates the wind-resistant design and structural safety. However, the underlying flow mechanisms in complex building clusters remain under-explored, mainly due to their nonlinear and configuration-dependent behavior. This study integrates wind tunnel experiments with an explainable artificial intelligence (XAI) framework to provide high-fidelity prediction and physical interpretation of aerodynamic interference within triangular high-rise clusters. Systematic experiments varying streamwise and transverse spacing and rotational angles produced detailed surface pressure datasets. Four AI models, i.e., Support Vector Regression, Decision Tree, Random Forest, and XGBoost, were trained to predict mean and fluctuating pressure coefficients, with XGBoost yielding the best overall performance. Model interpretability, achieved through SHapley Additive exPlanations (SHAP), revealed that transverse spacing governs regime transitions between shielding and resonance amplification, while streamwise spacing primarily influences fluctuating pressures through aerodynamic damping. SHAP analysis also identified pronounced three-dimensional pressure non-uniformity and a rotation-induced converging nozzle effect that increases mean pressures while moderating fluctuations. The proposed XAI-assisted framework establishes a data-driven approach for uncovering aerodynamic interference mechanisms, thus providing insights for resilient and performance-informed wind design of high-rise building clusters.