Conditional Neural Network Wind Pressure Statistical Estimation on Building Structures

This publication provides a set of neural network models for the estimation of wind pressure distribution over building surfaces. The models have two inputs, i.e., the wind condition vector and the coordinates vector. The wind condition vector has 6 elements [width, depth, height, roof slope, incidence wind angle, side number]. The second input is the normalized location coordinate vector on the surface of the building, which are the actual coordinates divided by the corresponding building dimensions. Researchers and engineers that need to know wind pressure distribution on low-rise buildings can use this model to estimate the target building wind load. Details about parameters, use case models, and how to run the models are located in the readme file. The models have been trained using data from the Tokyo Polytechnic University (TPU) low-rise building wind tunnel test database. Transfer learning can be used to accommodate other buildings.