摘要
针对电力负荷序列时序动态性、多尺度特征及复杂周期规律给预测带来的难题,提出一种基于时频双域特征融合与动态交互机制的短期电力负荷预测方法,其核心架构为双谱网。首先,针对短期电力负荷数据的非平稳和非线性特性,采用基于阿尔法进化算法改进的变分模态分解算法对负荷数据分解,得到若干本征模态函数;其次,设计频域特征增强机制,通过频谱注意力动态融合振幅谱与相位谱,并构建时频交叉注意力网络嵌入频域先验,结合跨维度门控实现特征校准;最后,基于多尺度金字塔解码器自适应融合时空特征生成预测值。以某市电力负荷数据集进行验证并与主流模型进行对比,结果表明所采用的预测方法具有更好的预测性能。
To address the challenges posed by temporal dynamics,multi-scale features,and complex periodic patterns in short-term power load forecasting,this paper proposes a short-term load forecasting(STLF)method based on dynamic time-frequency fusion,whose core architecture is DSN.First,to handle the non-stationary and non-linear characteristics of short-term power load data,an improved variational mode decomposition(VMD)algorithm,optimized by an adaptive evolution algorithm,is employed to decompose the load data into several intrinsic mode functions.Second,a frequency domain feature enhancement mechanism is designed,dynamically fusing amplitude and phase spectra through spectral attention.A time-frequency cross-attention network is then constructed to embed frequency domain priors,combined with a cross-dimensional gating mechanism for feature calibration.Finally,a multi-scale pyramid decoder is utilized to adaptively fuse the spatio-temporal features and generate the final predictions.Experimental results on a real-world load dataset from a city in China demonstrate that the proposed method achieves superior forecasting performance compared to mainstream models.
作者
王东风
张浩
胡怡然
崔玉雷
黄宇
WANG Dongfeng;ZHANG Hao;HU Yiran;CUI Yulei;HUANG Yu(Department of Automation,North China Electric Power University,Baoding 071003,China)
出处
《电力科学与工程》
2025年第12期57-64,共8页
Electric Power Science and Engineering
基金
国家自然科学基金资助项目(52207235)。
关键词
时频双域
动态交互
双谱网
频域特征增强
多尺度金字塔解码器
time-frequency dual-domain
dynamic interaction
dual spectral network(DSN)
frequency-domain feature enhancement
multi-scale pyramid decoder
