Page 37 - 《中国电力》2026年第5期

第 59 卷 第 5 期 Vol. 59, No. 5
2026 年 5 月 ELECTRIC POWER May 2026

引用格式：张怀天, 贾东梨, 王帅, 等. 基于 Transformer-集成学习的配电网短期负荷预测方法[J]. 中国电力, 2026, 59(5): 33−45.
Citation: ZHANG Huaitian, JIA Dongli, WANG Shuai, et al. Short-term load forecasting method for distribution networks based on transformer and
ensemble learning[J]. Electric Power, 2026, 59(5): 33−45.

基于 Transformer-集成学习的配电网

短期负荷预测方法

张怀天 ，贾东梨 ，王帅 ，何开元 ，任昭颖 ，刘佳静 ，胡雪凯 2
1
1
1
1
1
1
(1. 中国电力科学研究院有限公司，北京 100192；2. 国网河北省电力有限公司，河北 石家庄 050011)
Short-term load forecasting method for distribution networks based on
transformer and ensemble learning
ZHANG Huaitian , JIA Dongli , WANG Shuai , HE Kaiyuan , REN Zhaoying , LIU Jiajing , HU Xuekai 2
1
1
1
1
1
1
(1. China Electric Power Research Institute, Beijing 100192, China; 2. State Grid Hebei Electric Power Co., Ltd., Shijiazhuang 050011, China)

Abstract: Against the backdrop of the new power systems, the This work is supported by Smart Grid-National Science and
penetration rate of distributed energy resources in distribution Technology Major Project (No.2025ZD0804600).
networks is rising steadily, and the load characteristics are Keywords: short-term load forecasting; Transformer; ensemble
becoming increasingly diversified. Existing short-term load learning; Dropout; forward propagation sampling
forecasting methods thus fail to effectively capture the high-
dimensional nonlinear temporal characteristics of load data. To 摘 要：在新型电力系统背景下，配电网分布式能源渗
address this issue, this paper proposes a short-term load 透率不断提高，负荷特性日趋多元，传统短期负荷预测
方法难以有效捕捉高维非线性时序特征。为此，提出一
forecasting method for distribution networks based on
种 基 于 Transformer-集 成 学 习 的 短 期 负 荷 预 测 方 法 。 首
Transformer and ensemble learning. First, a multi-dimensional
先，构建多维特征嵌入层，融合负荷时序、周期特征及
feature embedding layer is constructed to fuse the temporal and
环境变量；其次，采用多头自注意力机制建立跨时段动
periodic characteristics of loads as well as environmental
态关联，提取负荷的时空耦合特性；然后，设计分层随
variables. Second, a multi-head self-attention mechanism is
机化前馈网络，结合 Dropout 增强模型隐空间的多模态表
adopted to establish dynamic cross-time interval correlations,
征能力；最后，集成多个差异化 Dropout 模型，通过多次
thereby extracting the spatiotemporal coupling characteristics of
前向传播采样实现对预测不确定性的贝叶斯评估。实验
loads accurately. Third, a hierarchical randomized feedforward
结果表明，所提方法在预测精度与稳定性上均优于现有
network is designed, with the Dropout technique integrated to
基准模型，可为配电网优化调度提供有效支持。
enhance the multimodal representation capability of the
关键词：短期负荷预测；Transformer；集成学习；Dropout
model’s latent space. Finally, multiple differentiated Dropout-
策略；前向传播采样
based models are ensembled, and Bayesian evaluation of
DOI：10.11930/j.issn.1004-9649.202511063

forecasting uncertainty is realized through sampling with
multiple forward propagations. Experimental results
1 引言
demonstrate that the proposed method outperforms state-of-the-
art benchmark models in both forecasting accuracy and
stability, and can thus provide effective technical support for 随着新型电力系统的建设与演进，由基础刚
the optimal dispatching of distribution networks. 性负荷、柔性可调负荷及随机波动性负荷构成的
多元负荷体系正在配电网中迅速增长。其中，照
收稿日期：2025−11−21； 修回日期：2026−04−26。 明、常规家电等基础刚性负荷保持其固有用电模
基金项目：智能电网重大专项（2030）资助项目（2025ZD0804 式；空调、电动汽车充电桩等柔性负荷具备一定
600）。 时空调节潜力；而规模化电动汽车充电与用户侧

33