基于变分模态分解-样本熵-延迟互信息-深度信念网络的短期负荷预测模型  

Short-Term Load Forecasting Model Based on Variational Modal Decomposition-Sample Entropy-Delay Mutual Information-Deep Belief Network

作　　者：徐轶丹 周晨梦 黄绍书 王枫 陈飞 胡洁 XU Yidan;ZHOU Chenmeng;HUANG Shaoshu;WANG Feng;CHEN Fei;HU Jie(DC company of State Grid Hubei Electric Power Co.,Ltd.,Yichang 443000,Hubei,China;Gui’an Power Supply Bureau of Guizhou Power Grid Co.,Ltd.,Gui’an 550025,Guizhou,China)

机构地区：[1]国网湖北省电力有限公司直流公司,湖北宜昌443000 [2]贵州电网有限责任公司贵安供电局,贵州贵安550025

出　　处：《电力大数据》2022年第10期10-20,共11页Power Systems and Big Data

摘　　要：传统的短期负荷预测模型未考虑组合预测模型的在数据处理上的优势,为了提高短期负荷预测的精度,本文提出了一种短期负荷预测的模型。首先,以降低电力负荷序列非平稳性的影响为目的,采用了变分模态分解方法将原始电力负荷序列分解为一系列的不同特征信息且相对较平稳的固有模态分量,分析掌握每个分量的变化规律;其次,利用样本熵理论对分解的各个分量进行复杂度分析以减少计算规模,提高预测模型的计算效率;然后考虑相关因素对负荷影响的延迟效应,采用了计及延迟效应的互信息特征选择技术重构原始输入序列;最后,结合深度信念网络预测模型,建立基于变分模态分解样本熵延迟效应互信息深度信念网络的组合预测模型,仿真结果表明该模型的有效性。The traditional short-term load forecasting model does not consider the advantages of combined forecasting model in data processing. In order to improve the accuracy of short-term load forecasting, this paper proposes a new short-term load forecasting model. Firstly, aimed at reducing the influence of non-stationarity of power load series, the variational modal decomposition method is used to decompose the original power load series into a series of relatively stable inherent modal components with different characteristic information, and analyze and master the variation law of each component. Secondly, the complexity of the decomposed components is analyzed by using the sample entropy theory to reduce the calculation scale and improve the calculation efficiency of the prediction model.Then considering the delay effect of related factors on the load, mutual information feature selection technology considering the delay effect is used to reconstruct the original input sequence.Finally, combined with the prediction model of deep belief network, a combined prediction model based on VMD-SE-DMI-DBN is established. The simulation results shows the validity of the proposed model.

关 键 词：负荷预测 互信息 模态分解 熵 负荷数据 

分 类 号：TM715[电气工程—电力系统及自动化]