| 引用本文: | 祝欣宇,窦迅,牛鹏艺,郭艳敏,石飞.基于改进ViT模型的电网关键线路智能预测方法[J].电力自动化设备,2026,46(2):205-214. |
| ZHU Xinyu,DOU Xun,NIU Pengyi,GUO Yanmin,SHI Fei.Improved ViT model based intelligent prediction method of key transmission lines in power grid[J].Electric Power Automation Equipment,2026,46(2):205-214. |
|
| 摘要: |
| 针对现有关键线路辨识方法在应对多源不确定性和复杂交易的准确性与适应性不足的问题，提出了一种基于改进视觉转换器(ViT)模型的电网关键线路智能预测方法。剖析了关键线路智能预测的原理，提出了考虑中长期交易和新能源不确定性的关键线路评价指标；采用多目标组合赋权方法，基于排序学习策略动态平衡主客观权重以优化排序目标；引入多尺度感知模块和上采样操作改进ViT模型，以增强对时序-指标数据的特征提取能力，通过通道扩展与空间适配机制提升其对全局依赖与多尺度特征的表征能力以实现关键线路预测。算例分析结果表明，该方法预测准确率达97.9 %，在中长期交易场景下具备良好的有效性与适应性。 |
| 关键词: 中长期交易 关键线路 智能预测 改进ViT模型 多尺度感知 |
| DOI：10.16081/j.epae.202508025 |
| 分类号: |
| 基金项目:国家电网有限公司科技项目(5108-202218280A-2-290-XG) |
|
| Improved ViT model based intelligent prediction method of key transmission lines in power grid |
|
ZHU Xinyu1, DOU Xun1, NIU Pengyi1, GUO Yanmin2, SHI Fei2
|
|
1.College of Electrical Engineering and Control Science, Nanjing TECH University, Nanjing 211816, China;2.China Electric Power Research Institute, Nanjing 210003, China
|
| Abstract: |
| Aiming at the insufficient accuracy and adaptability of identifying key transmission lines under multi-source uncertainty and accuracy of complex transactions, an improved vision transformer(ViT) model based intelligent prediction method of key transmission lines in power grid is proposed. The prediction principles of key transmission lines are analyzed, the evaluation indices of key transmission lines considering the medium and long-term transaction and renewable energy uncertainty are constructed. A multi-objective combined weighting method is employed to optimize the ranking target, in which the subjective and objective weights are dynamically balanced via a learning-to-rank strategy. The ViT model is enhanced with a multi-scale perception module and up-sampling operations to improve the ability of feature extraction from time-indexed data. Global dependencies and multi-scale representational capability are strengthened by channel expansion and spatial adaptation mechanism. Case studies demonstrate that the proposed method achieves a prediction accuracy of 97.9 %,verifying its effectiveness and applicability in medium and long-term transaction scenarios. |
| Key words: medium and long-term transaction key transmission line intelligent prediction improved ViT model multi-scale perception |
