Ensemble power load forecasting based on competitive-inhibition selection strategy and deep learning

Short-term power load forecasting plays a crucial role in improving the operational efficiency and economic benefit of the state grid system. To enhance the power load forecasting performance, several models have been proposed, and various combined models have shown good performance. Nevertheless, many current combined models neglect the importance and necessity of data pretreatment; in particular, they do not fully consider different noise patterns in different datasets. Moreover, they rarely consider deep learning in their model combinations. The current research gap limits the forecasting enhancement of the current combined model. In this paper, an ensemble power load forecasting system, which combines a competitive-inhibition feature selection strategy and a deep-learning-participating model, is proposed to address this gap. This ensemble forecasting system successfully improves the forecasting efficiency, and a case study with real 30-min power load data from four cities in Australia clearly demonstrates that the proposed system is significantly better than the comparison models. Therefore, the proposed system is a valid tool for smart grid planning.