Short-Term Power Load Forecasting Based on Wavelet Transform and Deep Deterministic Policy Gradient

The accurate prediction of power load is of great significance for the safe operation of the smart grid and the power transactions of market participants since the power load data series exhibits nonlinearity and volatility. In this paper, a new hybrid approach for deterministic short-term power load forecasting is proposed based on wavelet transform and deep deterministic policy gradient. In this approach, the original load data sequence is first decomposed by wavelet transform into some sub-frequency load sequences, and each sub-frequency can have better outlines and behavior. A deep deterministic policy gradient model is then employed to extract nonlinear features and invariant structures of each sub-frequency for power load. Finally, a new reward function for imbalanced samples is developed to effectively evaluate the policy score of the actor-network and further improve the prediction performance of the deep deterministic policy gradient. The proposed deterministic forecasting approach is used to actual power load data from an independent system operator from a city in China. The prediction results of the proposed method are presented in case studies, which have been demonstrated to achieve superior performance in terms of seasons and various prediction horizons compared with other prediction models.