An algorithm for forecasting day-ahead wind power via novel long short-term memory and wind power ramp events

Reliable wind power and ramp event prediction is essential for the safe and stable operation of electric power systems. Previous prediction methods struggled to forecast large fluctuations in wind power caused by extreme weather conditions, severely limiting the development of wind power prediction techniques. Based on this problem, an improved hybrid model is presented in this study, that utilises long short-term memory (LSTM) by considering wind power ramp events (WPREs). First, the LSTM network was driven by numerical weather prediction (NWP) to forecast day-ahead wind power. Second, a novel improved dynamic swinging door algo-rithm (ImDSDA) and a fuzzy C-means (FCM) model were utilised for WPRE detection and classification respectively. Third, a similarity-matching mechanism was proposed to correct the predicted WPREs. Finally, the predicted wind power was reconstructed using the optimised WPREs.The model, which was validated in three mountainous wind farms in central China, can capture the temporal dynamics of wind power using deep learning and WPRE prediction. The proposed model's results outperformed a few existing methods and can provide scientific guidance for the safe dispatching and economic operation of power systems.

Automated summary

This study proposes an improved hybrid model that utilizes long short-term memory and wind power ramp event prediction to forecast day-ahead wind power fluctuations. The model outperforms existing methods and provides guidance for the safe dispatching and economic operation of power systems.