Page 146 - 《中国电力》2026年第5期
P. 146

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

              引用格式:杨超颖, 李慧蓬, 赵军. 基于深度学习的集中式光伏电站在线增量功率预测方法[J]. 中国电力, 2026, 59(5): 142−149.
              Citation: YANG Chaoying, LI Huipeng, ZHAO Jun. Online incremental power forecasting method for centralized photovoltaic power plants based
              on deep learning[J]. Electric Power, 2026, 59(5): 142−149.



                             基于深度学习的集中式光伏电站在线

                                              增量功率预测方法



                                               杨超颖,李慧蓬,赵军

                                     (国网山西省电力公司电力科学研究院,山西 太原 030001)

                  Online incremental power forecasting method for centralized photovoltaic

                                        power plants based on deep learning
                                           YANG Chaoying, LI Huipeng, ZHAO Jun
                           (Electric Power Research Institute, State Grid Shanxi Electric Power Co., Ltd., Taiyuan 030001, China)

              Abstract:  Under  the  "dual  carbon"  goals,  centralized  photo-  摘 要:在“双碳”目标下,集中式光伏电站已成为新
              voltaic (PV) power stations have become a crucial support for  能源电力系统的重要支撑,但光伏发电功率受季节、天
              the  new  energy  power  syst  however,  PV  power  generation  is  气等因素影响具有强间歇性与波动性。针对实际应用中
              strongly intermittent and volatile due to factors such as seasons  输入数据动态演变导致模型性能衰减,且传统更新方式
              and  weather.  Addressing  the  issue  of  model  performance  易引发灾难性遗忘的问题,提出一种基于深度学习的在
              degradation caused by the dynamic evolution of input data in  线增量功率预测模型。该模型引入深度经验回放++(deep
              practical acations, and the susceptibility of traditional updating  experience replay,DER++)增量学习机制,构建“分块特
              methods to catastrophic forgetting, this paper proposes a deep  征提取-在线知识保留”双核心框架,通过补丁令牌策略
              learning-based online incremental power prediction model. The  捕捉多尺度周期性特征,利用自注意力机制挖掘多变量
              model  introduces  the  Deep  Experience  Replay  (DER)  依赖关系,结合经验回放技术缓解灾难性遗忘。基于某
              incremenng mechanism to construct a dual-core framework of  光伏电站实测数据表明,所提模型的累计精度衰减率远
              "block  feature  extraction  and  online  knowledge  retention".  It  低于传统模型,展现出更强的适应性与泛化能力,为集
              captures multi-scale periodic features via a patch token strategy,  中式光伏功率在线动态预测提供了有效解决方案。
              utilizes  self-attention  mechanto  mine  multivariate  depend-  关键词:集中式光伏电站;功率预测;深度学习
              encies, and combines experience replay techniques to alleviate  DOI:10.11930/j.issn.1004-9649.202510014

              catastrophic  forgetting.  Experimental  results  based  on  real-
              world data from a PV power station indicate that the cumulative  0    引言
              accuracy  degradation  rate  of  the  prsed  model  is  significantly
              lower  than  that  of  traditional  models,  demonstrating  stronger
                                                                    在全球“双碳”目标推进与能源结构转型的
              adaptability  and  generalization  capabilities,  and  providing  an
                                                                大背景下,集中式光伏电站凭借装机容量大、能
              effective  solution  for  online  dynamic  power  prediction  in
                                                                源输出稳定、并网效率高等优势,已成为中国新
              centralized.
                                                                能源电力系统的重要电源支撑                [1-6] 。实现集中式光
              This work is supported by Science and Technology Project of
                                                                伏 电 站 功 率 的 高 精 度 预 测 , 尤 其 是 在 线 动 态 预
              State Grid Shanxi Electric Power Co., Ltd. (No.52053023000M).
              Keywords:  centralized  photovoltaic  power  station;  power  测,已成为保障新能源电力系统安全、经济、高
              prediction; deep learning                         效运行的关键技术需求            [7-13] 。
                                                                    针对集中式光伏功率预测问题,国内外学者
              收稿日期:2025−10−10; 修回日期:2026−04−17。                 已开展了大量研究,形成了基于物理模型、统计
              基金项目:国网山西省电力公司科技项目(52053023000M)。                 模型与机器学习模型等主流预测方法。物理模型

              142
   141   142   143   144   145   146   147   148   149   150   151