军事知识服务平台
基于多维注意力Transformer与时序提示词的极端风速下风电出力概率预测方法
Probabilistic Forecasting Method for Wind Power Output in Extreme Wind Speed Scenarios Based on Multi-dimensional Attention Transformer and Temporal Prompting
摘要 在新型电力系统高比例风电接入与能源转型加速推进的背景下,风电出力固有的强非平稳性、间歇性及极端天气诱发的非线性突变特征,给电力系统的实时平衡与安全调度带来严峻挑战.针对极端风速下传统数据驱动深度学习方法难以有效捕获风机保护停机引发的预测区间偏移与分布失配问题,提出一种基于多维注意力Transformer与时序提示词的极端风速下风电出力概率预测方法.该方法通过构建协同的时间与协变量维度双重注意力机制,深度融合并提取风电序列的多维时空演化规律;同时创新性地设计面向极端工况的时序提示词机制,将外部停机物理先验信息转化为模型可理解的语义引导向量,实现对预测分布的自适应动态修正.仿真分析表明,所提模型能够有效克服极端工况下的预测失配难题,显著提升概率预测的可靠性与锐度,为新型电力系统在复杂气象条件下的安全稳定运行提供了关键理论支撑与技术手段.
Abstract:
Against the backdrop of large-scale wind power integration into new-type power systems and the accelerating energy transition,the inherent strong non-stationarity,intermittency,and nonlinear abrupt changes induced by extreme weather conditions in wind power output pose unprecedented and severe challenges to real-time power balance and secure dispatch of power systems.To address the issues that traditional data-driven deep learning methods fail to effectively capture the prediction interval shift and distribution mismatch caused by wind turbine protective shutdowns under extreme wind speed scenarios,this paper proposes a probabilistic forecasting method for wind power output in such scenarios based on a multi-dimensional attention Transformer and temporal prompting.The proposed method constructs a collaborative dual-attention mechanism along both the temporal and covariate dimensions,enabling deep fusion and extraction of the multi-dimensional spatio-temporal evolution patterns inherent in wind power sequences.Meanwhile,it innovatively designs a temporal prompting mechanism tailored for extreme operating conditions,which transforms external physical prior information of shutdown events into semantically guided vectors comprehensible to the model,thereby achieving adaptive and dynamic correction of the predictive distribution.Simulation results demonstrate that the proposed model can effectively overcome the forecasting mismatch problem under extreme conditions,significantly enhancing both the reliability and sharpness of probabilistic predictions,thus providing critical theoretical support and technical means for the safe and stable operation of new-type power systems under complex meteorological conditions.
刊 名 智慧电力 ISTICEIPKU
年,卷(期): 2026, 54(9)
分类号: TM614
机标分类号: TM711 TP391 TM614
在线出版日期: 2026年11月9日