摘要
针对智慧楼宇负荷类型复杂且多变导致的负荷预测精度低等问题,提出了一种基于改进递归网络的智慧楼宇负荷预测方法.该方法在深度神经网络多隐层结构的基础上增设了关联层,使深度递归神经网络(DRNN)模型具有动态特性,并利用改进粒子群优化算法对模型权值空间进行优化,进而实现楼宇负荷的准确预测.基于不同类型楼宇的实验结果表明,所提方法的预测误差约在±0.3 MW的范围内波动,其均方根差与平均绝对百分比误差分别为0.27 MW和1.05%,且预测误差均小于其他对比方法.
Aiming at the problem of low load forecasting accuracy caused by complex and changeable load types of intelligent buildings,an intelligent building load forecasting method based on improved recursive network was proposed.Based on the multi hidden layer structure of deep neural network,an association layer was added to make the deep recurrent neural network(DRNN)model exhibit dynamic characteristics.An improved particle swarm optimization algorithm was used to optimize the model weight space,thereby realizing the accurate forecasting of building load.The experimental results on different types of buildings show that the prediction error of as-proposed method fluctuates in the range of±0.3 MW,and the root mean square error and mean absolute percentage error are 0.27 MW and 1.05%,respectively.The prediction errors are smaller than those of other analogue methods.
作者
肖荣洋
黄雁
XIAO Rong-yang;HUANG Yan(School of Electronic and Information Engineering,Tongji University,Shanghai 200331,China;Longyan Power Supply Company,State Grid Fujian Electric Power Co.Ltd.,Longyan 364000,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2022年第2期121-126,共6页
Journal of Shenyang University of Technology
基金
福建省自然科学基金项目(2018J01746)
国网福建省电力有限公司龙岩供电公司项目(SGFJLY00YJJS2100564).
关键词
改进粒子群优化算法
深度递归神经网络
智慧楼宇
负荷预测
关联层
均方根差
平均绝对百分比误差
预测误差
improved particle swarm optimization algorithm
deep recursive neural network
intelligent building
load forecasting
correlation layer
root mean square error
mean absolute percentage error
prediction error
