Short‐term load forecasting utilizing wavelet transform and time series considering accuracy feedback

Summary The short‐term electric load forecasting (STLF) is of great significance for the secure and economic operation of power systems. Due to the integration of distributed generation and spreading of demand response programs, uncertainty of electric load has been greatly aggravated. Consequently, maintaining the accuracy of STLF has become a challenging task. Moreover, for traditional STLF studies, the prediction errors are usually utilized for result evaluation and comparison but have not been fully embedded into the established models. In this paper, a hybrid model named as wavelet time series (WTS) is proposed, which integrates Daubechies wavelet transformer (WT), Holt‐Winters (HW) and radical basis function neural network (RBF) to achieve rapid calculation and a satisfactory nonlinear load fitting ability. The WT is used to decompose the load series into a set of frequency components and each component is separately forecasted. Two alternative historical load and weather datasets with different temporal scales are established. The HW and RBF models are both implemented on the same days of interest with two different kinds of historical input. A result‐driven accuracy feedback mechanism is introduced to adaptively determine the optimal forecasting model for each decomposed load series on each specific day. It is verified by the numerical results on a realistic Chinese load dataset that the proposed WTS model is capable to improve the short‐term forecasting accuracy of several load indices simultaneously.