Forecasting of Tropical Storm Wind Speeds Based on Multi-Step Differencing and Artificial Neural Network

The tropical storm is a severe wind disaster that frequently attacks coastal structures and infrastructure facilities. Accurate wind speed forecasting of tropical storms, based on real-time measured data, has become a critical issue in the engineering community. Utilizing the measured data of typical tropical storms at Sutong Bridge, this study develops a new approach for wind speed forecasting, which integrates multi-step differencing with an artificial neural network-based model. Given the non-stationary nature of tropical storm wind speeds, a multi-step differencing operation is initially applied to the wind speed time series. Subsequently, multi-step predictions of the differenced wind speeds are made for future time points. Finally, an inverse differencing operation is employed to reconstruct the wind speeds to be forecasted. The forecasting errors associated with single-step differencing, multi-step differencing, and no differencing are compared to evaluate their respective performances. To validate the generalizability of the developed approach, it is further used in the wind speed forecasting of another typhoon wind speed dataset. The satisfactory performance demonstrates the effectiveness of the developed approach for multi-step wind speed forecasting of tropical storms.