基于RF的并行CNN-TGLSTM热负荷预测模型  

RF-based Parallel CNN-TGLSTM Model for Heat Load Prediction

作　　者：谭全伟 薛贵军 谢文举 TAN Quan-wei;XUE Gui-jun;XIE wen-ju(College of Electrical Engineering,North China University of Science and Technology,Tangshan Hebeoi 063210,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063210

出　　处：《华北理工大学学报（自然科学版）》2024年第2期112-123,共12页Journal of North China University of Science and Technology：Natural Science Edition

基　　金：河北省自然科学基金项目(E2020209121)。

摘　　要：精准的热负荷预测不仅可以提高用户舒适度,还可以有效降低能源消耗。为了提升热负荷预测的准确性,本研究提出了一种基于随机森林的并行CNN和TGLSTM的短期热负荷预测模型。首先,采用随机森林算法对特征进行筛选;其次,利用并行网络CNN和改进的LSTM分别提取时空特征;最后,将提取的特征与多头注意力机制动态结合。实验结果表明,并行CNN-TGLSTM-MA相较于传统的串行CNN-TGLSTM模型,在MAE和MSE方面分别降低了16.9%、26.8%,同时提升了3.5%的R2值,证明了所提出的并行CNN-TGLSTM-MA模型在短期热负荷预测方面的有效性和优越性,为热力系统供热负荷的精准调控提供了参考。Accurate heat load forecasting can not only improve user comfort,but also effectively reduce energy consumption.In order to improve the accuracy of heat load prediction,a short-term heat load prediction model based on random forest and parallel CNN-TGLSTM attention was proposed.Firstly,the random forest algorithm was used to carry out feature engineering on the data.Secondly,parallel network CNN and improved LSTM were used to extract spatiotemporal features respectively.Finally,the extracted features were dynamically combined with the multi-head attention mechanism.The experimental results show that compared with the traditional serial CNN-TGLSTM-MA model,the MAE and MSE of the parallel CNN-TGLSTM model are reduced by 16.9%and 26.8%,respectively,and the R~2 value is increased by 3.5%.The validity and superiority of the proposed parallel CNN-TGLSTM-MA model in short-term heat load forecasting are proved.It provides a reference for the precise control of heating load in thermal system.

关 键 词：短期热负荷预测 卷积神经网络 转换门控长短期记忆网络 多头注意力机制 随机森林 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]