基于 TimesNet-Transformer 融合模型的电能质量时序预测方法

危康耀; 郝瑗宇; 杨冬冬; 李丹; 王嘉乐; 郑丽君

doi:10.19783/j.cnki.pspc.260161

欢迎来到能源电力期刊网！

English Version 期刊登录

您当前的位置：

首页 >

文章列表页 >

基于 TimesNet-Transformer 融合模型的电能质量时序预测方法

理论分析 | 更新时间：2026-07-27

- 基于 TimesNet-Transformer 融合模型的电能质量时序预测方法
- A power quality time-series forecasting method based on the TimesNet-Transformer fusion model
- 电力系统保护与控制 2026年54卷第13期
- 作者机构：
- 作者简介：
- 基金信息：
- DOI：10.19783/j.cnki.pspc.260161
  中图分类号：
- 纸质出版：2026
- 稿件说明：
移动端阅览
危康耀, 郝瑗宇, 杨冬冬, 等. 基于 TimesNet-Transformer 融合模型的电能质量时序预测方法[J]. 电力系统保护与控制, 2026,54(13).

WEI Kangyao, HAO Yuanyu, YANG Dongdong, et al. A power quality time-series forecasting method based on the TimesNet-Transformer fusion model[J]. Power System Protection and Control, 2026, 54(13).
危康耀, 郝瑗宇, 杨冬冬, 等. 基于 TimesNet-Transformer 融合模型的电能质量时序预测方法[J]. 电力系统保护与控制, 2026,54(13). DOI： 10.19783/j.cnki.pspc.260161.

WEI Kangyao, HAO Yuanyu, YANG Dongdong, et al. A power quality time-series forecasting method based on the TimesNet-Transformer fusion model[J]. Power System Protection and Control, 2026, 54(13). DOI： 10.19783/j.cnki.pspc.260161.

摘要

为解决传统深度学习模型难以充分挖掘电能质量数据中复杂时间特征的问题，提出一种基于 TimesNet-Transformer 融合模型的预测方法。该方法首先利用 TimesNet 中的时序二维变换模块，将一维时间序列转换为二维空间结构，提取复杂时间特征。其次，引入多层 Transformer 网络，捕捉特征中的长期依赖关系。然后，设计自适应加权融合模块，对两个分支所提取的特征进行对齐与加权融合，强化关键信息，并通过全连接层输出最终的预测结果。最后，以山西省太原市某地区电网全年实际运行数据构建数据集进行验证。实验结果表明，所提方法在模型计算精度上较 Transformer 和 TimesNet 分别提升 11.83% 和 3.16%，训练效率提升约 29.32%。欧洲公开数据集跨域泛化验证显示，融合模型在核心电能质量指标上均取得最优精度，验证了其在复杂电网场景下的适应性与鲁棒性。

Abstract

To address the challenges that traditional deep learning models struggle to fully extract the complex temporal features from power quality data

a forecasting method based on the TimesNet-Transformer hybrid model is proposed. The proposed method first uses the temporal two-dimensional transformation module in TimesNet to convert one-dimensional time-series data into a two-dimensional spatial structure for extracting complex temporal feature vectors. Second

a multi-layer Transformer network is introduced to capture long-term dependencies within these features. Then

an adaptive weighted fusion module is designed to align and integrate the features extracted by the two branches

thereby enhancing critical information representation and generating the final forecasting results through a fully connected layer. Finally

a dataset constructed from a full year of actual operational data collected from a regional power grid in Taiyuan

Shanxi Province

is used for validation. Experimental results show that the proposed method achieves an average improvement of 11.83% and 3.16% in model computation accuracy compared with Transformer and TimesNet

respectively

while increasing training efficiency by approximately 29.32%. Cross-domain generalization validation on a European public dataset further demonstrates that the hybrid model achieves the highest prediction accuracy across key power quality indicators

confirming its adaptability and robustness under complex power grid scenarios.

关键词

Keywords

references

TOMAS V, STANISLAV M, TOMAS J, et al. The power quality forecasting model for off-grid system supported by multiobjective optimization[J]. IEEE Transactions on Industrial Electronics, 2017, 64(12): 9507-9516.

ZHANG C, WANG Y, FU Y, et al. A novel DW TimesNet-based short-term multi-step wind power forecasting model using feature selection and auto-tuning methods[J]. Energy Conversion and Management, 2024, 301.

HE Jun, YUAN Kuidong, ZHONG Zijie, et al. Enhancing short-term power load forecasting with a TimesNet-Crossformer-LSTM Approach[J]. IEEE Access, 2024, 12: 56774-56788.

NADA M, HAMID O, ISMAEL J. Short-term electric load forecasting using an EMD-BI-LSTM approach for smart grid energy management system[J]. Energy Buildings, 2023, 288.

ZHANG Zhe, WANG Bo, YANG Mao, et al. Short-term prediction of wind farm cluster power based on bidirectional interactive fusion of homogeneous and heterogeneous information[J]. Protection and Control of Modern Power Systems, 2026, 11(2): 107-125.

MOHANASUNDARAM V, RANGASWAMY B. Elastic net with Bayesian density estimation model for feature selection for photovoltaic energy prediction[J]. Scientific Reports, 2025, 15(1): 8736-8746.

VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C] // Advances in Neural Information Processing Systems, December 4-9, 2017, Long Beach, CA, USA: 5998-6008.

ZHOU H, ZHANG S, PENG J, et al. Informer: beyond efficient transformer for long sequence time-series forecasting[C] // Proceedings of the AAAI Conference on Artificial Intelligence, February 2-9, 2021: 11106-11115.

WU H, XU J, WANG J, et al. Autoformer: decomposition transformers with auto-correlation for long-term series forecasting[C] // Advances in Neural Information Processing Systems, December 6-14, 2021: 22418-22430.

ZHOU T, MA Z, WEN Q, et al. FEDformer: frequency enhanced decomposed transformer for long-term series forecasting[C] // Proceedings of the 39th International Conference on Machine Learning, July 17-23, 2022, Baltimore, Maryland, USA: 27268-27286.

ZENG A, CHEN M, ZHANG L, et al. Are transformers effective for time series forecasting?[C] // Proceedings of the AAAI Conference on Artificial Intelligence, February 7-14, 2023, Washington, DC, USA: 11121-11128.

LIN S, LIN W, WU W, et al. SegRNN: segment recurrent neural network for long-term time series forecasting[C]// The Twelfth International Conference on Learning Representations, May 7-11, 2024, Vienna, Austria.

WU H, HU T, LIU Y, et al. TimesNet: temporal 2D-variation modeling for general time series analysis[C] // The Eleventh International Conference on Learning Representations, May 1-5, 2023, Kigali, Rwanda: 1-6.

LADISLAV Z. Power quality multi-step predictions with the gradually increasing selected input parameters using machine-learning and regression[J]. Sustainable Energy, Grids and Networks, 2021, 100.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

考虑多频次耦合影响的配电网谐波电流估计方法

基于 CONMI-LSGC-Transformer 的串联电弧故障检测

计及预案式失配冲击的响应驱动频率稳定紧急切负荷策略

注入式与非注入式协同的逆变器阻抗估计

融合功率变化率与裕度反馈的 VSC-MTDC 自适应下垂控制