摘要
针对虚拟电厂负荷特性日益复杂、预测难度加大等问题,本文提出了一种基于改进变分模态分解(Variational Mode Decomposition,VMD)与深度时序动态网络的多元负荷预测方法。首先,采用冠豪猪优化算法(Crested Porcupine Optimizer,CPO)对VMD进行优化,对输入特征进行分解,有效提升了模型的稳定性。其次,通过分析多元负荷的耦合特性,结合季节性负荷预测方法,利用时序卷积网络(Temporal Convolutional Network,TCN)精准捕捉负荷数据中的长期时间依赖特征。同时,引入双向门控循环单元(Bidirectional Gate Recurrent Unit,BiGRU),显著增强了模型对时间序列双向动态特性的捕获能力。随后,结合注意力机制对关键时间步进行聚焦,进一步强化了模型对关键特征的感知能力及对局部与全局信息的综合理解能力。最后,通过算例分析验证,本文所提多元负荷预测模型相较于单一预测模型及传统预测模型,在运算效率和预测精确度上均表现出显著优势。
In response to the increasingly complex load characteristics and heightened prediction challenges of virtual power plant,this paper proposes a multivariate load forecasting method based on an improved Variational Mode Decomposition(VMD)and a deep time series dynamic network.Firstly,the Crested Porcupine Optimizer(CPO)algorithm is employed to optimize VMD,decomposing input features and effectively enhancing the model s stability.Secondly,by analyzing the coupling characteristics of multivariate loads and integrating seasonal load forecasting methods,Temporal Convolutional Networks(TCN)are utilized to precisely capture long-term temporal dependencies in load data.Simultaneously,the introduction of Bidirectional Gated Recurrent Units(BiGRU)significantly enhances the model s ability to capture bidirectional dynamic characteristics of time series.Subsequently,an attention mechanism is incorporated to focus on critical time steps,further strengthening the model s perception of key features and its comprehensive understanding of both local and global information.Finally,through case study analysis,the proposed multivariate load forecasting method demonstrates significant advantages in both computational efficiency and prediction accuracy compared to single forecasting models and traditional forecasting models.
作者
胡伟
庞代宇
杜璞良
HU Wei;PANG Daiyu;DU Puliang(School of Economics and Management,Shanghai University of Electric Power,Shanghai 201306,China)
出处
《复旦学报(自然科学版)》
北大核心
2025年第4期395-411,共17页
Journal of Fudan University:Natural Science
基金
国家自然科学基金(72401182)。
关键词
虚拟电厂
多元负荷预测
冠豪猪优化算法
变分模态分解
深度时序动态网络
virtual power plant
multivariate load forecasting
crested porcupine optimizer algorithm
variational mode decomposition
deep time series dynamic network
作者简介
胡伟(1979-),男,博士,教授;通信作者:庞代宇,女,硕士研究生,E-mail:pangdaiyu1211@163.com。
