PhaCIA−TCNs: Short−Term Load Forecasting Using Temporal Convolutional Networks With Parallel Hybrid Activated Convolution and Input Attention
Zhenghua Xu‚ Zhoutao Yu‚ Hexiang Zhang‚ Junyang Chen‚ Junhua Gu‚ Thomas Lukasiewicz and Victor Leung
Short-term load forecasting (STLF) is an important task in modern smart grids, which reduces energy waste and enhances the reliability of the power system. RNN-based models are the most widely used deep-learning-based methods for STLF, but their forecasting performance is limited due to an unbalanced non-linearity problem. Temporal convolution networks (TCNs) are thus proposed to remedy this problem, however, TCNs have two shortcomings, i.e., redundant convolutional operation and equal input importance problems. Therefore, we propose a novel TCN-based backbone model, called PhaCIA-TCNs, to achieve a more accurate short-term load forecasting, where parallel hybrid activated convolution (PhaC) and input attention (IA) are proposed to resolve the above problems of TCNs. PhaC shortens the convolutional learning path and time-cost of TCNs while maintaining superior feature learning capabilities, and IA is used to highlight important input elements while depress irrelevant ones. Extensive experimental results show that (i) PhaCIA-TCNs significantly outperform all state-of-the-art RNN-based and TCNs-based baselines in forecasting-error-based evaluation metrics on all datasets; (ii) PhaC and IA are both effective and essential for PhaCIA-TCN to achieve the superior STLF performances; and (iii) despite achieving better forecasting accuracies, the training time-cost of PhaCIA-TCN is similar (and sometimes even lower) to the existing TCNs-based solution.