Short-term load forecasting based on sample weights assignment

Abstract

Short-term load forecasting (STLF) is the basis of the power system operation. Considering that the importance of different training samples is different, a sample weights assignment method is proposed in this paper to help the STLF to learn the key sample. At first, the sample similarity is measured considering the characteristics of different input components. Based on this, training samples are selected. Finally, different training samples are assigned with different sample weights through the designed weights assignment function. With the proposed method, the STLF model is able to focus on the crucial samples. Simulation results considering different data-driven models demonstrate the effectiveness of the proposed method.

Keywords

• Short-term load forecasting (STLF)
• Sample similarity measurement
• Training samples selection
• Sample weights assignment
• Data-driven STLF model