摘要
针对综合能源系统(integrated energy system,IES)中因多元负荷复杂性和耦合性导致的预测精度受限问题,提出一种基于滚动模态分解和GCN-DABiLSTM的IES多元负荷预测模型。首先,利用完全自适应噪声集合经验模态分解(complete ensemble empirical mode decomposition with adaptive noise,CEEMDAN)对电、冷、热负荷进行初步分解,生成一系列子序列;其次,采用模糊散布熵(fuzzy discrete entropy,FDE)对子序列进行复杂性评估并聚合;然后,通过变分模态分解(variational mode decomposition,VMD)对高频分量进行二次分解,将原始序列解耦为特征聚焦且平稳的子序列。在分解过程中,引入滚动分解策略,规避了传统基于模态分解的预测方法带来的信息渗透问题。最后,构建一个结合图卷积网络(graph convolutional network,GCN)和双重注意力(dual-attention mechanism,DA)机制优化的双向长短期记忆网络(bidirectional long short-term memory,BiLSTM)的组合预测框架,用于多元负荷预测。基于美国亚利桑那州立大学IES数据的验证表明,该模型相较于其他模型,预测误差显著降低,验证了其在预测任务中的有效性。
To address the issue of limited prediction accuracy in the integrated energy system(IES)due to the complexity and coupling of multiple loads,this paper proposes a multi-load prediction model for IES based on rolling mode decomposition and GCN-DABiLSTM.Firstly,the complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)is utilized to conduct an initial decomposition of electricity,cooling,and heating loads,generating a series of subsequences.Subsequently,fuzzy discrete entropy(FDE)is employed to evaluate the complexity of the subsequences and aggregate them.Then,variational mode decomposition(VMD)is applied to perform a secondary decomposition of the high-frequency components,decoupling the original sequence into feature-focused and stationary subsequences.During the decomposition process,a rolling decomposition strategy is introduced to avoid the information leakage problem brought by traditional mode decomposition-based prediction methods.Finally,a combined prediction framework integrating the graph convolutional network(GCN)and dual-attention mechanism(DA)optimized bidirectional long short-term memory(BiLSTM)is constructed for multi-load prediction.The validation based on the integrated energy system data from Arizona State University in the United States shows that this model significantly reduces the prediction errors compared to other models,verifying its effectiveness in prediction.
作者
罗林霖
王霄
何志琴
尹曜华
LUO Linlin;WANG Xiao;HE Zhiqin;YIN Yaohua(The College of Electrical Engineering,Guizhou University,Guiyang,Guizhou 550025,China;Power China Guiyang Engineering Co.,Ltd.,Guiyang,Guizhou 550081,China)
出处
《广东电力》
北大核心
2025年第9期130-144,共15页
Guangdong Electric Power
基金
中国电力建设股份有限公司科技项目(DJ-ZDXM-2022-44)
关键词
负荷预测
滚动模态分解
模糊散布熵
图卷积网络
双向长短期记忆网络
双重注意力机制
load forecasting
rolling mode decomposition
fuzzy discrete entropy(FDE)
graph convolutional network(GCN)
bidirectional long short-term memory(BiLSTM)
dual-attention mechanism(DA)
