Artificial neural networks for predicting mean wind profiles over heterogeneous terrains

This paper presents the application of artificial neural networks (ANNs) for predicting mean wind profile characteristics over heterogeneous terrains. The ANN models integrate salient terrain features to predict the vertical wind profile structure of neutrally stable atmospheric boundary layer (ABL) flows. The research leveraged wind profile data collected in a large boundary layer wind tunnel equipped with a mechanized roughness element grid, which enabled the simulation of a wide range of heterogeneous terrain conditions. Three ANN architectures are examined to determine the most critical terrain features that influence wind profile prediction. Specifically, several input parameters are investigated to capture proximate and distal roughness changes upstream to the measurement location. The results demonstrate the efficacy of the proposed ANN-based approach in accurately predicting mean wind profiles over heterogeneous terrains. While the ANN models exhibit a higher degree of accuracy and reliability, they require large volumes of data that may not be easily accessible. However, the research findings will help advance predictive modeling in wind engineering and deepen our understanding of boundary layer physics by identifying key parameters and developing strategies to accurately capture wind profiles over complex heterogeneous terrains. •Three multi-layer perceptron artificial neural networks (ANNs) are designed to predict the mean profile structure of wind fields over heterogeneous terrains.•The ANN models were trained using velocity profile data collected in a large boundary layer wind tunnel (BLWT) at University of Florida (UF).•60 real-world heterogenous sites were simulated in the UF BLWT using a mechanized roughness element grid with an 18-m long fetch.•Several input features are investigated to capture proximate and distal roughness changes upstream to the measurement location.•ANN model predictions reveal a high degree of accuracy and reliability, demonstrating its potential as a valuable tool for wind engineering.