STUDY OF A METHOD FOR PREDICTING INPUT WIND SPEED TIME HISTORY FOR A DOME-SHAPED WIND TUNNEL USING LSTM MODEL

Abstract

This study investigated a method for obtaining the input wind speed time history to reproduce the target wind speed time history in a dome-shaped wind tunnel using a machine learning model. In particular, in this report, the effects of various parameters used to set up the LSTM model on the prediction accuracy were investigated when the input wind speed is set at only one surface and the output wind speed is set at only the horizontal wind direction component. Then, the CFD input wind speed to reproduce the target wind speed was predicted using the machine learning model, and the CFD output wind speed obtained with the CFD input wind speed was compared with the target wind speed. The results showed that the prediction accuracy was relatively good for the parts of the time history where the wind speed changes slowly, but tended to be less accurate for the parts where the wind speed changes rapidly in a short period of time and for the peaks and troughs of the time history.