渐进式分层特征提取的综合能源多任务负荷预测

PDF下载HTML

Title:

Multi-task load forecasting of integrated energy based on progressive layered feature extraction

作者:

王德文^1,2, 安涵¹, 张林飞¹, 赵文清^1,3

1. 华北电力大学 计算机系, 河北 保定 071003;
2. 华北电力大学 河北省能源电力知识计算重点实验室, 河北 保定 071003;
3. 华北电力大学 复杂能源系统智能计算教育部工程研究中心, 河北 保定 071003

Author(s):

WANG Dewen^1,2, AN Han¹, ZHANG Linfei¹, ZHAO Wenqing^1,3

1. Department of Computer, North China Electric Power University, Baoding 071003, China;
2. Hebei Key Laboratory of Knowledge Computing for Energy & Power, North China Electric Power University, Baoding 071003, China;
3. Engineering Research Center of Intelligent Computing for Complex Energy Systems, Ministry of Education, North China Electric Power University, Baoding 071003, China

关键词:

负荷预测; 综合能源; 多任务学习; 多元负荷; 渐进式分层; 特征提取; 最大信息系数; 变分模态分解

Keywords:

load forecasting; integrated energy; multi-task learning; multiple loads; progressive layered; feature extraction; maximum information coefficient; variational mode decomposition

分类号:

TP18

DOI:

10.11992/tis.202406008

文献标志码:

2025-2-25

摘要:

针对综合能源系统中电、冷、热负荷存在复杂耦合关系，传统多任务学习模型难以学习到有效的多元负荷耦合特征可能导致预测精度降低的问题，本文充分考虑多元负荷复杂耦合关系，提出一种渐进式分层特征提取的综合能源多任务负荷预测模型。将全年数据按季节划分，分析各季节下电、冷、热负荷间耦合强度；采用变分模态分解将历史负荷序列分解为多个不同频率的分量，可以更好挖掘多元负荷的深层时序特征；渐进式分层提取多元负荷的耦合特征，并动态分配耦合特征对预测结果的影响权重，避免耦合特征无效时模型预测精度下降。实验结果证明，在不同的多元负荷耦合强度下，渐进式分层特征提取的多任务负荷预测在精度上有更好表现。研究结论可用于指导综合能源多元负荷预测过程。

Abstract:

Due to the complex coupling relationship between electric, cold and heat loads in an integrated energy system, it is difficult for traditional multi-task learning models to learn effective multi-load coupling characteristics, which may lead to reduced prediction accuracy. In this paper, a comprehensive energy multi-task load forecasting model with progressive layered feature extraction is proposed, considering the complex coupling relationship of multiple loads. Firstly, divide the annual data by season and analyze the coupling strength between electricity, cooling, and heating loads in each season. Then, by using variational mode decomposition, the historical load sequence is decomposed into multiple components of different frequencies, which can better explore the deep time series features of multiple loads. Finally, the coupling features of multiple loads are extracted progressively and the influence weights of the coupling features on the prediction results are dynamically allocated to avoid the degradation of the model prediction accuracy when the coupling features are invalid. Experimental results show that the proposed model has better performance in terms of prediction accuracy under different multi-component load coupling intensities. The conclusion can be used to guide the process of load forecasting of integrated energy.