基于自适应Transformer的短期负荷预测域适应方法  

A Study on the Domain Adaptation Method for Short-Term Load Forecasting Based on the Self-Adaptive Transformer

作　　者：周俊[1] 马泽菊[1] ZHOU Jun;MA Zeju(Chongqing Electric Power College,Chongqing 400053,P.R.China)

机构地区：[1]重庆电力高等专科学校,重庆400053

出　　处：《重庆电力高等专科学校学报》2025年第1期1-6,共6页Journal of Chongqing Electric Power College

基　　金：重庆市教委科技项目(KJZD-K202102601)。

摘　　要：有效的STLF对于当今电力系统的平稳运行至关重要。然而,传统的预测方法往往无法应对电力消费数据中的复杂性、非线性和动态变化,特别是在处理具有不同数据分布的新区域时。为了克服这些限制,提出了针对STLF的ATDA。ATDA利用Transformer编码器有效的建模时间依赖性,并结合带有重要性加权的部分对抗域适应策略,解决源域和目标域之间的差异。通过优先考虑与目标域最相关的源样本,ATDA最小化了负迁移并提高了预测精度。在来自国家电网公司的真实数据上进行的综合实验表明,ATDA在预测性能上显著优于当前领先的模型。An effective STLF is essential for the smooth operation of the power system.However,traditional forecasting methods often fail to cope with the complexity,non-linearity and dynamic changes in electricity consumption data,especially when dealing with new domains with different data distributions.To overcome these limitations,this paper proposes an ATDA for the STLF,which uses the Transformer encoder to effectively model time dependence and resolve differences between source and target domains in combination with the partial adversarial domain adaptation strategy with importance weighting.By prioritizing source samples that are most relevant to the target domain,the ATDA minimizes negative migrations and improves the forecasting accuracy.Comprehensive experiments on real data from State Grid Corporation of China show that the ATDA significantly outperforms current leading models in forecasting performance.

关 键 词：短期负荷预测 域适应 TRANSFORMER 对抗学习 重要性加权 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]