Long short-term memory (LSTM)-based wind speed prediction during a typhoon for bridge traffic control

A short-term wind speed prediction framework is proposed for bridge traffic control under strong winds. The framework mainly focuses on improving the prediction accuracy for the timeframe of traffic control during a typhoon. Two concepts are newly proposed to achieve the goal: 1) hybrid modeling of wind speed at the bridge; and, 2) the adoption of a time-shifted data correction (TSDC) method. First, the hybrid modeling considers two available data types, one from a structural health monitoring system of the bridge and the other from the regional specialized meteorological center (RSMC). The training features of a long short-term memory (LSTM) approach are chosen based on the maximum sustained winds of a typhoon. Second, the TSDC method accounts for a time delay phenomenon between the maximum wind speed at the bridge deck and the maxima or minima of the selected features. The Mean Absolute Error (MAE)-based grid search method determines the preferable combinations of two parameters: input data length and the time-shifted length of the training data. As a numerical example, typhoons from 2020 are used as test data to demonstrate the improvement in prediction performance via the use of hybrid modeling and the TSDC method.

Automated summary

A framework for short-term wind speed prediction in bridge traffic control under strong winds is proposed, aiming to improve prediction accuracy. The framework incorporates hybrid modeling and a time-shifted data correction method.