[Paper Review] Investigation of wind pressures on tall building under interference effects using machine learning techniques

Gang Hu, Lingbo Liu+3 more|arXiv (Cornell University)|Aug 20, 2019

Wind and Air Flow Studies57 references4 citations

TL;DR

This study employs generative adversarial networks (GANs) to predict wind pressure coefficients on tall buildings under interference effects using only 30% of a comprehensive wind tunnel dataset from Tokyo Polytechnic University. The GANs model trained on 30% of the data outperformed other machine learning models and accurately predicted both mean and fluctuating pressure coefficients under unseen interference conditions, enabling up to 70% reduction in required wind tunnel tests.

ABSTRACT

Interference effects of tall buildings have attracted numerous studies due to the boom of clusters of tall buildings in megacities. To fully understand the interference effects of buildings, it often requires a substantial amount of wind tunnel tests. Limited wind tunnel tests that only cover part of interference scenarios are unable to fully reveal the interference effects. This study used machine learning techniques to resolve the conflicting requirement between limited wind tunnel tests that produce unreliable results and a completed investigation of the interference effects that is costly and time-consuming. Four machine learning models including decision tree, random forest, XGBoost, generative adversarial networks (GANs), were trained based on 30% of a dataset to predict both mean and fluctuating pressure coefficients on the principal building. The GANs model exhibited the best performance in predicting these pressure coefficients. A number of GANs models were then trained based on different portions of the dataset ranging from 10% to 90%. It was found that the GANs model based on 30% of the dataset is capable of predicting both mean and fluctuating pressure coefficients under unseen interference conditions accurately. By using this GANs model, 70% of the wind tunnel test cases can be saved, largely alleviating the cost of this kind of wind tunnel testing study.

Motivation & Objective

• To address the high cost and limited scope of wind tunnel testing for studying interference effects in clusters of tall buildings.
• To develop a machine learning model capable of accurately predicting wind pressures under unseen interference conditions with minimal experimental data.
• To evaluate and compare the performance of multiple machine learning models—decision tree, random forest, XGBoost, and GANs—for predicting wind pressure coefficients.
• To determine the minimum dataset size required for a GANs model to achieve reliable predictions across diverse interference scenarios.
• To enable high-resolution mapping of interference factors for structural design without exhaustive physical testing.

Proposed method

• Trained four machine learning models—decision tree, random forest, XGBoost, and generative adversarial networks (GANs)—on 30% of a dataset from the Tokyo Polytechnic University aerodynamic database.
• Used a GANs architecture to learn the underlying distribution of wind pressure coefficient data across various building arrangements, wind directions, and spacing ratios.
• Evaluated model performance using statistical metrics such as R² and RMSE between predicted and experimental pressure coefficients.
• Trained multiple GANs models on increasing dataset portions (10% to 90%) to assess data efficiency and model generalization.
• Validated the best-performing GANs model against unseen test cases to confirm its predictive accuracy for untested interference configurations.
• Leveraged the trained GANs model to generate predictions for 70% of the original wind tunnel test cases, effectively replacing physical testing.

Experimental results

Research questions

• RQ1 Can machine learning models trained on a limited subset of wind tunnel test data accurately predict wind pressure coefficients on tall buildings under interference effects?
• RQ2 Which machine learning model—decision tree, random forest, XGBoost, or GANs—performs best in predicting both mean and fluctuating pressure coefficients under unseen interference conditions?
• RQ3 What is the minimum dataset size required for a GANs model to achieve reliable and generalizable predictions across diverse building arrangements and wind conditions?
• RQ4 To what extent can a trained GANs model replace physical wind tunnel testing in interference effect studies, and what is the potential reduction in required test cases?
• RQ5 Can the trained GANs model generate high-resolution interference factor maps for force and moment coefficients that are otherwise infeasible through conventional testing?

Key findings

• The GANs model outperformed decision tree, random forest, and XGBoost in predicting both mean and fluctuating pressure coefficients across all tested conditions.
• A GANs model trained on just 30% of the full 2,664-case dataset achieved prediction accuracy comparable to models trained on the entire dataset.
• Performance improvements from increasing training data were most significant below 30%, with diminishing returns beyond that threshold.
• The GANs model trained on 30% of the data accurately predicted pressure coefficients under unseen interference conditions, including critical locations with high dynamic amplification.
• The model enabled the prediction of 70% of wind tunnel test cases without physical testing, significantly reducing cost and time in wind engineering studies.
• The study demonstrates that a GANs model trained on a small data subset can be more effective than a model trained on the full dataset, due to better generalization and noise suppression.