Page 5 - 《中国电力》2026年第5期
P. 5

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

              引用格式:陈梓宏, 黄宁馨, 赖智航, 等. 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.
              Citation: CHEN Zihong, HUANG Ningxin, LAI Zhihang, et al. 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.


                      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 ,黄宁馨 ,赖智航 ,赖晓文 ,陈潇婷 ,陈硕楠 ,高锋                                                  3
                                                                    2
                                                                                             2
                                                                                2
                              1
                                          1
                                                       1
                   (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)
                   An improved Transformer day-ahead electricity price forecasting model
                         based on Patch mechanism and channel-independent structure

                                                               1
                                                                            1
                                              1
                                                                                          2
                                 CHEN Zihong , HUANG Ningxin , LAI Zhihang , LAI Xiaowen ,
                                         CHEN Xiaoting , CHEN Shuonan , GAO Feng   3
                                                                       2
                                                       2
              (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)

              Abstract:  To  address  the  common  problems  of  insufficient  This work is supported by National Natural Science Foundation
              temporal feature extraction and poor adaptability to special day  of China (No.72401011).
              scenarios in day-ahead electricity spot market price forecasting,  Keywords: electricity spot market; day-ahead price forecasting;
              this paper proposes an improved forecasting model based on the  channel-independent structure; multi-head attention mechanism
              Transformer  architecture.  The  Patch  mechanism  is  introduced
              to enhance local temporal feature extraction, and the channel-  摘 要:针对电力现货市场日前电价预测中普遍存在的
              independent  structure  is  combined  to  improve  the  learning  时 序 特 征 提 取 不 足 、 特 殊 日 类 型 场 景 适 应 性 差 的 问 题 ,
              efficiency of multivariate features. In addition, the multi-head  提 出 一 种 基 于  Transformer 架 构 的 改 进 预 测 模 型 。 引 入
              attention  mechanism  is  adopted  to  capture  the  global  price  Patch  机制增强局部时序特征提取,结合通道独立结构增
              fluctuation patterns. The proposed method is verified based on  加多变量特征学习效率,通过多头注意力机制捕获全局
              the  historical  data  of  the  Guangdong  electricity  spot  market.  电价波动规律。基于广东省电力现货市场历史数据进行
              Compared  with  the  baseline  Transformer  model,  the  mean  方法验证,与基准  Transformer 模型相比,周末场景的平
              absolute error (MAE) of the proposed model is decreased from  均绝对误差从  32.95  降低至 23.88,节假日场景的平均绝对
              32.95 to 23.88 in weekend scenarios, and from 78.33 to 70.33  误差从  78.33  降低至  70.33。对量价偏移现象的适应性显著
              in  holiday  scenarios.  The  model  exhibits  significantly  better  优于基准模型,在竞价空间大于  6  万  MW  时能准确捕捉价
              adaptability to the phenomenon of quantity-price deviation than  格下限上升趋势,所提方法在不同场景(特别是特殊场
              the baseline model, and can accurately capture the upward trend  景)预测精度显著提升,对量价偏移现象适应性好。
              of  price  floors  when  the  bidding  space  exceeds  60 000  MW.  关 键 词 : 电 力 现 货 市 场 ; 日 前 电 价 预 测 ; 通 道 独 立 结
              The  proposed  model  achieves  a  significant  improvement  in  构;多头注意力机制
              prediction accuracy under different scenarios (especially special  DOI:10.11930/j.issn.1004-9649.202506072

              scenarios)  and  has  good  adaptability  to  quantity-price
              deviations.                                       0    引言


              收稿日期:2025−06−30; 修回日期:2026−03−28。                     随 着 全 球 能 源 结 构 转 型 和 电 力 市 场 逐 步 开
              基金项目:国家自然科学基金资助项目(72401011)。                      放,电价预测已成为电力系统经济运行和市场运

                                                                                                            1
   1   2   3   4   5   6   7   8   9   10