一种基于Patch机制与通道独立结构的改进Transformer日前电价预测方法

doi:10.11930/j.issn.1004-9649.202506072

中国电力 ›› 2026, Vol. 59 ›› Issue (5): 1-8.DOI: 10.11930/j.issn.1004-9649.202506072

• “双碳”目标下支撑能源转型的电价市场化改革与价格监管 • 上一篇下一篇

一种基于Patch机制与通道独立结构的改进Transformer日前电价预测方法

陈梓宏¹(), 黄宁馨¹(), 赖智航¹, 赖晓文², 陈潇婷², 陈硕楠²(), 高锋³

1. 广东粤电电力销售有限公司，广东广州　510630
2. 北京清能互联科技有限公司，北京　100084
3. 北京工业大学，北京　100124

收稿日期:2025-06-30 修回日期:2026-03-28 发布日期:2026-05-15 出版日期:2026-05-28
作者简介:
陈梓宏（1993），男，工程师，从事电力市场、电力经济研究，E-mail：chenzihong@geg.com.cn
黄宁馨（1997），女，硕士研究生，从事电力经济、电力供需研究，E-mail：huangningxin@geg.com.cn
陈硕楠（1999），男，通信作者，硕士研究生，从事人工智能、机器学习算法应用研究，E-mail：1025877071@qq.com
基金资助:
国家自然科学基金资助项目（72401011）。

An improved Transformer day-ahead electricity price forecasting model based on Patch mechanism and channel-independent structure

CHEN Zihong¹(), HUANG Ningxin¹(), LAI Zhihang¹, LAI Xiaowen², CHEN Xiaoting², CHEN Shuonan²(), GAO Feng³

1. Guangdong Yudean Electric Marketing Co., Ltd., Guangzhou 510630, China
2. Beijing TsIntergy Technology Co., Ltd., Beijing 100084, China
3. Beijing University of Technology, Beijing 100124, China

Received:2025-06-30 Revised:2026-03-28 Online:2026-05-15 Published:2026-05-28
Supported by:
This work is supported by National Natural Science Foundation of China (No.72401011).

摘要/Abstract

摘要：

针对电力现货市场日前电价预测中普遍存在的时序特征提取不足、特殊日类型场景适应性差的问题，提出一种基于Transformer架构的改进预测模型。引入Patch机制增强局部时序特征提取，结合通道独立结构增加多变量特征学习效率，通过多头注意力机制捕获全局电价波动规律。基于广东省电力现货市场历史数据进行方法验证，与基准Transformer模型相比，周末场景的平均绝对误差从32.95降低至 23.88，节假日场景的平均绝对误差从78.33降低至70.33。对量价偏移现象的适应性显著优于基准模型，在竞价空间大于6万MW时能准确捕捉价格下限上升趋势，所提方法在不同场景（特别是特殊场景）预测精度显著提升，对量价偏移现象适应性好。

关键词: 电力现货市场, 日前电价预测, 通道独立结构, 多头注意力机制

Abstract:

To address the common problems of insufficient temporal feature extraction and poor adaptability to special day scenarios in day-ahead electricity spot market price forecasting, this paper proposes an improved forecasting model based on the Transformer architecture. The Patch mechanism is introduced to enhance local temporal feature extraction, and the channel-independent structure is combined to improve the learning efficiency of multivariate features. In addition, the multi-head attention mechanism is adopted to capture the global price fluctuation patterns. The proposed method is verified based on the historical data of the Guangdong electricity spot market. Compared with the baseline Transformer model, the mean absolute error (MAE) of the proposed model is decreased from 32.95 to 23.88 in weekend scenarios, and from 78.33 to 70.33 in holiday scenarios. The model exhibits significantly better adaptability to the phenomenon of quantity-price deviation than the baseline model, and can accurately capture the upward trend of price floors when the bidding space exceeds 60000 MW. The proposed model achieves a significant improvement in prediction accuracy under different scenarios (especially special scenarios) and has good adaptability to quantity-price deviations.

Key words: electricity spot market, day-ahead price forecasting, channel-independent structure, multi-head attention mechanism

陈梓宏, 黄宁馨, 赖智航, 赖晓文, 陈潇婷, 陈硕楠, 高锋. 一种基于Patch机制与通道独立结构的改进Transformer日前电价预测方法[J]. 中国电力, 2026, 59(5): 1-8.

CHEN Zihong, HUANG Ningxin, LAI Zhihang, LAI Xiaowen, CHEN Xiaoting, CHEN Shuonan, GAO Feng. An improved Transformer day-ahead electricity price forecasting model based on Patch mechanism and channel-independent structure[J]. Electric Power, 2026, 59(5): 1-8.

导出引用管理器 EndNote|Ris|BibTeX

链接本文: https://www.electricpower.com.cn/CN/10.11930/j.issn.1004-9649.202506072

https://www.electricpower.com.cn/CN/Y2026/V59/I5/1

图/表 4

参考文献 33

1	黄日星, 康重庆, 夏清. 电力市场中的边际电价预测[J]. 电力系统自动化, 2000, 24 (22): 9- 12, 51.
	HUANG Rixing, KANG Chongqing, XIA Qing. System marginal price forecasting in electricity market[J]. Automation of Electric Power Systems, 2000, 24 (22): 9- 12, 51.
2	王耀华, 李晓纯, 杨宇, 等. 全球能源转型与地缘环境的交互关系[J]. 中国电力, 2025, 58 (12): 1- 13.
	WANG Yaohua, LI Xiaochun, YANG Yu, et al. Interactions between global energy transition and geopolitical environment[J]. Electric Power, 2025, 58 (12): 1- 13.
3	石榴, 张乾, 王建光, 等. 火电机组在电力现货市场中的最优报价策略模型研究[J]. 电工技术, 2025 (10): 226- 229.
	SHI Liu, ZHANG Qian, WANG Jianguang, et al. Research on the optimal pricing strategy model of thermal power units in the spot electricity market[J]. Electric Engineering, 2025 (10): 226- 229.
4	CANTILLO-LUNA S, MORENO-CHUQUEN R, LOPEZ-SOTELO J, et al. An intra-day electricity price forecasting based on a probabilistic transformer neural network architecture[J]. Energies, 2023, 16 (19): 6767.
5	张静, 熊国江. 考虑季节特性与数据窗口的短期光伏功率预测组合模型[J]. 电力工程技术, 2025, 44 (1): 183- 192.
	ZHANG Jing, XIONG Guojiang. Short-term photovoltaic power prediction combination model considering seasonal characteristic and data window[J]. Electric Power Engineering Technology, 2025, 44 (1): 183- 192.
6	KOLB S, DILLIG M, PLANKENBÜHLER T, et al. The impact of renewables on electricity prices in Germany - An update for the years 2014–2018[J]. Renewable and Sustainable Energy Reviews, 2020, 134, 110307.
7	张进斌, 赵光泽, 高凤仓, 等. 基于IBA-LSTM的新能源日前电价预测方法研究[J]. 电力系统装备, 2025 (6): 73- 75.
	ZHANG Jinbin, ZHAO Guangze, GAO Fengcang, et al. Research on new energy day ahead electricity price prediction method based on IBA-LSTM[J]. Electric Power System Equipment, 2025 (6): 73- 75.
8	李艳梅, 顾诚凯, 任恒君, 等. 考虑决策者时长偏好的分时电价时段划分[J]. 电力工程技术, 2025, 44 (3): 131- 139.
	LI Yanmei, GU Chengkai, REN Hengjun, et al. Time-of-day division of time-of-use electricity tariff considering duration preference of the policymaker[J]. Electric Power Engineering Technology, 2025, 44 (3): 131- 139.
9	米静. 基于竞价算法的能源市场优化风电交易研究[J]. 电气应用, 2025, 44 (2): 73- 77.
	MI Jing. Research on energy market optimization of wind power trading based on bidding algorithm[J]. Electrotechnical Application, 2025, 44 (2): 73- 77.
10	舒征宇, 王喜召, 董超, 等. 含储能参与的日前市场价值公平分配机制[J]. 电力工程技术, 2024, 43 (2): 229- 238.
	SHU Zhengyu, WANG Xizhao, DONG Chao, et al. Mechanism for fair distribution of day-ahead market value with energy storage participation[J]. Electric Power Engineering Technology, 2024, 43 (2): 229- 238.
11	BOZLAK Ç B, YAŞAR C F. An optimized deep learning approach for forecasting day-ahead electricity prices[J]. Electric Power Systems Research, 2024, 229, 110129.
12	赵欣. 基于注意力机制的CNN-BiLSTM日前电价预测[J]. 电工技术, 2024 (12): 135- 138, 142.
	ZHAO Xin. CNN-BiLSTM day-ahead electricity price prediction based on attention mechanism[J]. Electric Engineering, 2024 (12): 135- 138, 142.
13	TAN Y Q, SHEN Y X, YU X Y, et al. Day-ahead electricity price forecasting employing a novel hybrid frame of deep learning methods: a case study in NSW, Australia[J]. Electric Power Systems Research, 2023, 220, 109300.
14	金亮, 尹振豪, 刘璐, 等. 基于残差U-Net和自注意力Transformer编码器的磁场预测方法[J]. 电工技术学报, 2024, 39 (10): 2937- 2952.
	JIN Liang, YIN Zhenhao, LIU Lu, et al. Magnetic field prediction method based on residual U-Net and self-attention transformer encoder[J]. Transactions of China Electrotechnical Society, 2024, 39 (10): 2937- 2952.
15	李璐, 阚小瑞, 毕贵红, 等. 基于注意力机制的LSTNet日前电价预测[J]. 电力科学与工程, 2025, 41 (4): 1- 10.
	LI Lu, KAN Xiaorui, BI Guihong, et al. LSTNet day-ahead electricity price prediction based on attention mechanism[J]. Electric Power Science and Engineering, 2025, 41 (4): 1- 10.
16	李练兵, 高国强, 吴伟强, 等. 考虑特征重组与改进Transformer的风电功率短期日前预测方法[J]. 电网技术, 2024, 48 (4): 1466- 1480.
	LI Lianbing, GAO Guoqiang, WU Weiqiang, et al. Short-term day-ahead wind power prediction considering feature recombination and improved transformer[J]. Power System Technology, 2024, 48 (4): 1466- 1480.
17	王慧强, 陈楚皓, 吕宏武, 等. 基于双向稀疏Transformer的多变量时序分类模型[J]. 小型微型计算机系统, 2024, 45 (3): 555- 561.
	WANG Huiqiang, CHEN Chuhao, LV Hongwu, et al. Multivariate time series classification model based on bidirectional sparse transformer[J]. Journal of Chinese Computer Systems, 2024, 45 (3): 555- 561.
18	LLORENTE O, PORTELA J. A transformer approach for electricity price forecasting[J/OL]. 2024: arXiv: 2403.16108[2025-03-30]. https://arxiv.org/abs/2403.16108.
19	王艳阳, 李建朝, 刘景青, 等. 电碳耦合机制下多元主体综合成本分析[J]. 中国电力, 2025, 58 (6): 180- 189.
	WANG Yanyang, LI Jianzhao, LIU Jingqing, et al. Comprehensive cost analysis of multiple entities under the coupling mechanism of electricity and carbon[J]. Electric Power, 2025, 58 (6): 180- 189.
20	李祥光, 谭青博, 李帆琪, 等. 电碳耦合对煤电机组现货市场结算电价影响分析模型[J]. 中国电力, 2024, 57 (5): 113- 125.
	LI Xiangguang, TAN Qingbo, LI Fanqi, et al. Analysis model to study the influence of electrocarbon coupling on settlement price of coal power units in spot market[J]. Electric Power, 2024, 57 (5): 113- 125.
21	刘超. 基于供需平衡的电力市场电价预测与调控研究[J]. 中国集体经济, 2024 (22): 41- 44.
	LIU Chao. Research on electricity price prediction and regulation in the power market based on supply and demand balance[J]. China Collective Economy, 2024 (22): 41- 44.
22	王蕾, 李斌, 吴飞, 等. 基于改进极限学习机的电力市场实时电价预测方法[J]. 电子设计工程, 2024, 32 (20): 21- 25, 30.
	WANG Lei, LI Bin, WU Fei, et al. Real time electricity price prediction method in the electricity market based on improved extreme learning machine[J]. Electronic Design Engineering, 2024, 32 (20): 21- 25, 30.
23	MOON H, CHUNG S, SUH J, et al. Analysis of power system flexibility considering power system ramp rate[J]. Journal of Electrical Engineering & Technology, 2024, 19 (2): 867- 876.
24	YAO X T, FU X L, ZONG C F. Short-term load forecasting method based on feature preference strategy and LightGBM-XGboost[J]. IEEE Access, 2022, 10, 75257- 75268.
25	XU C, CHEN S L, KONG X J, et al. Robust feature selection by removing noise entropy within mutual information for limited-sample industrial data[J]. IEEE Transactions on Industrial Informatics, 2025, 21 (5): 3913- 3923.
26	UNIEJEWSKI B, NOWOTARSKI J, WERON R. Automated variable selection and shrinkage for day-ahead electricity price forecasting[J]. Energies, 2016, 9 (8): 621.
27	LÖTSCH J, ULTSCH A. Recursive computed ABC (cABC) analysis as a precise method for reducing machine learning based feature sets to their minimum informative size[J]. Scientific Reports, 2023, 13, 5470.
28	徐宇颂, 邹山花, 卢先领. 基于特征选择和组合模型的短期电力负荷预测[J]. 中国电力, 2022, 55 (7): 121- 127.
	XU Yusong, ZOU Shanhua, LU Xianling. Short-term load forecasting based on feature selection and combination model[J]. Electric Power, 2022, 55 (7): 121- 127.
29	徐先峰, 赵依, 刘状壮, 等. 用于短期电力负荷预测的日负荷特性分类及特征集重构策略[J]. 电网技术, 2022, 46 (4): 1548- 1556.
	XU Xianfeng, ZHAO Yi, LIU Zhuangzhuang, et al. Daily load characteristic classification and feature set reconstruction strategy for short-term power load forecasting[J]. Power System Technology, 2022, 46 (4): 1548- 1556.
30	NIE Y Q, NGUYEN N H, SINTHONG P, et al. A time series is worth 64 words: long-term forecasting with transformers[J/OL]. 2022: arXiv: 2211.14730[2025-03-30]. https://arxiv.org/abs/2211.14730.
31	KAZEMI S M, GOEL R, EGHBALI S, et al. Time2Vec: learning a vector representation of time[J/OL]. 2019: arXiv: 1907.05321[2025-03-30]. https://arxiv.org/abs/1907.05321.
32	杨楠, 毕贵红, 李玉洪, 等. 基于Time2Vec-BiGRU-SA深度学习模型的碳价格预测[J]. 电力科学与工程, 2025, 41 (9): 1- 12.
	YANG Nan, BI Guihong, LI Yuhong, et al. Carbon price prediction based on Time2Vec-BiGRU-SA deep learning model[J]. Electric Power Science and Engineering, 2025, 41 (9): 1- 12.
33	李滨, 黄佳, 吴茵, 等. 基于分形特性修正气象相似日的节假日短期负荷预测方法[J]. 电网技术, 2017, 41 (6): 1949- 1955.
	LI Bin, HUANG Jia, WU Yin, et al. Holiday short-term load forecasting based on fractal characteristic modified meteorological similar day[J]. Power System Technology, 2017, 41 (6): 1949- 1955.

自注意力头数量H	编码维度$ {D}_{{\rm model}} $	测试集MAE	训练集MAE
4	96	52.25	43.24
	128	49.37	44.37
	256	50.25	38.58
	512	38.01	35.96
	768	44.55	45.90
6	96	50.55	49.99
	128	—	—
	256	46.22	43.07
	512	39.61	41.08
	768	34.94	32.62
8	96	48.03	42.80
	128	—	—
	256	41.40	41.60
	512	32.52	31.69
	768	43.38	39.78

自注意力头数量H	编码维度$ {D}_{{\rm model}} $	测试集MAE	训练集MAE
4	96	52.25	43.24
	128	49.37	44.37
	256	50.25	38.58
	512	38.01	35.96
	768	44.55	45.90
6	96	50.55	49.99
	128	—	—
	256	46.22	43.07
	512	39.61	41.08
	768	34.94	32.62
8	96	48.03	42.80
	128	—	—
	256	41.40	41.60
	512	32.52	31.69
	768	43.38	39.78

日类型	模型	数据集（节点A）		数据集（节点B)
日类型	模型	RMSE	MAE	RMSE	MAE
周末	A	40.81	32.95	50.93	45.78
	B	39.43	28.27	45.75	33.46
	C	40.32	30.90	47.12	36.41
	D	34.04	23.88	37.78	27.77
	XGBoost	53.75	46.17	67.21	56.15
工作日	A	71.74	46.53	80.21	59.31
	B	70.42	43.42	82.97	53.86
	C	70.63	43.74	76.67	52.57
	D	69.32	41.57	73.27	48.21
	XGBoost	88.61	49.27	92.67	62.83
节假日	A	137.37	78.33	155.82	92.18
	B	136.23	75.78	147.20	89.15
	C	129.90	75.52	138.51	87.01
	D	127.45	70.33	135.77	81.60
	XGBoost	122.67	68.89	131.39	78.86

日类型	模型	数据集（节点A）		数据集（节点B)
日类型	模型	RMSE	MAE	RMSE	MAE
周末	A	40.81	32.95	50.93	45.78
	B	39.43	28.27	45.75	33.46
	C	40.32	30.90	47.12	36.41
	D	34.04	23.88	37.78	27.77
	XGBoost	53.75	46.17	67.21	56.15
工作日	A	71.74	46.53	80.21	59.31
	B	70.42	43.42	82.97	53.86
	C	70.63	43.74	76.67	52.57
	D	69.32	41.57	73.27	48.21
	XGBoost	88.61	49.27	92.67	62.83
节假日	A	137.37	78.33	155.82	92.18
	B	136.23	75.78	147.20	89.15
	C	129.90	75.52	138.51	87.01
	D	127.45	70.33	135.77	81.60
	XGBoost	122.67	68.89	131.39	78.86

[1]	宋卓然, 张燕妮, 王阳, 蒋海玮, 李嘉宇, 高洪超. 电力市场环境下分布式能源系统的多元运行模式与灵活性影响分析[J]. 中国电力, 2025, 58(11): 111-121.
[2]	尚静怡, 张艺涵, 杨福旺, 向明旭. 发电商策略性竞价对省间-省内现货市场影响与应对策略[J]. 中国电力, 2025, 58(11): 1-13.
[3]	许凌, 张希鹏, 曹益奇, 张丙金, 董成, 谭振飞. 考虑备用互济的省间现货电能与备用耦合出清模型[J]. 中国电力, 2023, 56(9): 48-56.
[4]	关立, 周蕾, 刘航航, 周新生, 陶玮, 赵梓州. 独立储能电站参与电力现货市场机制及试运行分析[J]. 中国电力, 2022, 55(10): 185-190.
[5]	胡朝阳, 冯冬涵, 滕晓毕, 曾佳妮. 关于浙江电力现货市场若干关键问题的思考[J]. 中国电力, 2020, 53(9): 55-59,70.
[6]	王勇, 游大宁, 房光华, 张国强, 王进, 匡洪辉. 山东电力现货市场机制设计与试运行分析[J]. 中国电力, 2020, 53(9): 38-46.
[7]	张粒子, 丛野, 陶文斌, 郑福康, 陈正飞. 电力现货市场环境下提供位置信号电价机制的协同作用[J]. 中国电力, 2020, 53(9): 9-18,46.
[8]	裴哲义, 郭国梁, 胡超凡. 富余可再生能源跨区电力现货交易的探索与实践[J]. 中国电力, 2018, 51(1): 16-21.

一种基于Patch机制与通道独立结构的改进Transformer日前电价预测方法

An improved Transformer day-ahead electricity price forecasting model based on Patch mechanism and channel-independent structure

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 4

参考文献 33

相关文章 8

编辑推荐

Metrics

模态框（Modal）标题

一种基于Patch机制与通道独立结构的改进Transformer日前电价预测方法

An improved Transformer day-ahead electricity price forecasting model based on Patch mechanism and channel-independent structure

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 4

参考文献 33

相关文章 8

编辑推荐

Metrics