A spatial transfer-based hybrid model for wind speed forecasting

Accurate wind speed forecasting is essential for optimizing energy dispatch and enhancing grid stability. This study presents a novel hybrid wind speed forecasting model (WSDTW-CLA), emphasizing the spatial transfer characteristics of wind speed while mitigating the inherent errors in existing models. The proposed method employs the Wind Speed Dynamic Time Warping (WSDTW) algorithm to align wind speed data from neighboring stations, effectively facilitating the capture of spatial transfer patterns during the preprocessing phase. This alignment generates a wind speed spatial matrix that incorporates future-relevant information, providing precise input for forecasting module. The model employs a hybrid neural network combining a convolutional neural network (CNN), a long short-term memory (LSTM) network, and an autoencoder (AE) to predict wind speeds by establishing feature connections from the preprocessed data. The performance of the WSDTW-CLA model is evaluated using seasonal datasets from March, June, September, and December in Yunnan Province, China. A multi-step comparative analysis involving seven established models and seven sub-models within the proposed framework demonstrates that the WSDTW-CLA model significantly outperforms other similar models, with all evaluation metrics showing improvements of over 30 %. This proposed method enhances the utilization of wind energy resources, thereby promoting the advancement of the wind power industry.