改进时空图卷积神经网络的多设备能耗预测
Multi-Equipment energy consumption prediction based on improved STGCN
蔡同尧 1曾献辉1
作者信息
- 1. 东华大学信息科学与技术学院,上海 201620
- 折叠
摘要
为帮助企业实现高能耗设备的节能,针对负荷预测的已有研究大多集中在单一设备且重点考虑数据的时间特征,缺乏对设备空间特征的分析.文中提出一种基于时空图卷积神经网络(STGCN)的多设备能耗预测算法.该方法采用切比雪夫图卷积(Cheb-GC)层提取空间特征和改进的门控时间卷积(GTCN)层提取时间特征,通过嵌入图邻接矩阵达到多设备联合预测的目的.结合某高能耗企业的真实数据,与历史平均回归模型(HA)、自回归积分滑动平均模型(ARIMA)、前馈神经网络(FNN)和门控循环单元(GRU)网络进行了对比实验,结果表明所提算法具备更优越的性能,预测结果能够满足实际应用要求.
Abstract
In order to help enterprises realize energy saving of high energy consumption equipment.Most existing researches on equipment load forecasting focus on single equipment and give priority to the time characteristics of data,but lack the analysis of spatial characteristics.This paper a multi-equipment energy consumption prediction method based on Spatio-Temporal Graph Convolution Neural Network(STGCN)is proposed.The Chebyshev Graph Convolution(Cheb-GC)layer and the improved Gated Temporal Convolu-tion Network(GTCN)are used to extract spatial features and temporal features.Through embedding graph adjacency matrix,this method achieves the purpose of multi-equipment joint prediction.Combined with the real data set,the proposed model is compared with the Historical Average(HA)regression model,Autore-gressive Integrated Moving Average(ARIMA)model,Feedforward Neural Network(FNN)and Gated Re-current Unit(GRU)network.The results show that this algorithm has better performance,and the predic-tion results meet the requirements of practical application.
关键词
多设备能耗预测/时空特征/图神经网络/数据挖掘/深度学习Key words
multi-equipment energy consumption prediction/Spatio-Temporal characteristics/Graph Neu-ral Network/Data Mining/deep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据