Wind Speed Prediction Based on Wavelet Decomposition and the Fractal-Based LSTM Method

Accurate wind speed prediction is essential for wind power generation and wind-resistant structural design. In this research, a short-term wind speed prediction model that combines the wavelet decomposition (WD) method and a fractal-based long short-term memory (LSTM) network is proposed. The fractal dimensions of the dataset at each level, which are decomposed using the WD method, are calculated using the box-counting method. With the dynamic learning rate in the loss function updated by fractal dimensions, the gradient-enhanced LSTM network is applied for wind speed prediction. Experimental wind speed data collected during wind field measurement experiments in Pingtan, Fujian Province, China, were used to validate the proposed wind speed prediction model. The predicted wind speed results at different considered time intervals are compared with those of the traditional LSTM method. The results suggest that the proposed method significantly improves the accuracy of wind speed forecasting.