摘要
短期供热负荷预测对促进节能减排以及建设智慧供热系统有着重要意义。由于热负荷数据呈现较强的非线性、滞后性和耦合性,常规预测模型难以取得令人满意的效果。针对传统的长短时记忆(LSTM)神经网络的研究,提出了1种以小波分解(WD)和卷积神经网络(CNN)为基础的混合神经网络模型。首先,采用WD将序列分解为不同的子序列;其次,采用CNN提取局部特征,对输入变量进行解耦合;然后,采用LSTM对时间子序列分别进行预测;最后,通过融合计算子序列预测结果得出预测序列。试验结果表明,与LSTM、CNN、CNN-LSTM等网络相比,WD-CNN-LSTM的平均绝对误差、决定系数、均方根误差等指标都更好,能够满足换热站对供热负荷高精度预测的要求。
Short-term heating load prediction is of great significance to promote energy conservation and emission reduction as well as to build a smart heating system.Due to the strong nonlinearity,lag and coupling of heat load data,conventional prediction models are difficult to achieve satisfactory results.A hybrid neural network model based on wavelet decomposition(WD)and convolutional neural network(CNN)is proposed for the study of conventional long short term memory(LSTM)neural network.Firstly,the sequence is decomposed into different subsequences by using WD;secondly,local features are extracted by using CNN to decouple the input variables;then,LSTM is used to predict the time subsequences separately;finally,the predicted sequences are derived by using the predicted results of the computed subsequences.The experimental results show that compared with LSTM,CNN and CNN-LSTM networks,WD-CNN-LSTM is better in terms of average absolute error,coefficient of determination,root mean square error and other indexes,and can meet the requirements of heat exchange stations for high-precision prediction of heating loads.
作者
张嘉益
薛贵军
ZHANG Jiayi;XUE Guijun(College of Electrical Engineering,Chongqing University,Chongqing 400044,China;Instrument Factory,North China University of Technology,Tangshan 063000,China)
出处
《自动化仪表》
CAS
2023年第5期63-68,73,共7页
Process Automation Instrumentation
关键词
智慧供热
供热负荷预测
小波分解
单支重构
卷积神经网络
长短时记忆神经网络
Smart heating
Heating load prediction
Wavelet decomposition(WD)
Single branch reconstruction
Convoluational neural network(CNN)
Long short term memory(LSTM)neural network
