The influence of variational mode decomposition on LSTM prediction accuracy - A case study with wind turbine power signals

Sustainable development of energy is regarded as a crucial solution to combat global challenges such as climate change and energy security. However, the sustainable harnessing of wind energy confronts randomness and volatility, rendering the precise prediction of wind power signals a crucial challenge in the management of power systems. To address these issues, this study employs the Variational Mode Decomposition (VMD) method and Long Short-Term Memory (LSTM) networks to comprehensively analyze the impact of the VMD method on the accuracy of wind power signal prediction using LSTM. The orthogonal experimental method is used to determine the key parameters of the VMD method. The results show that the VMD-LSTM model significantly outperforms the traditional LSTM model in time series signal prediction accuracy. Compared to the LSTM model, the VMD-LSTM model reduces the MAE, MAPE, and RMSE errors by 39.7 %, 37.5 %, and 30.9 %, respectively, and increases the R2 by 20.6 %. Among the decomposed modes, the third-order mode exhibits the largest prediction error. A method is proposed to improve prediction accuracy by excluding certain non-essential modes. This study provides valuable insights into reducing power fluctuations in the grid caused by active power prediction errors.