基于深度学习的区域供热逐时负荷预测研究  

Study on Hourly Load Prediction of District Heating Based on Deep Learning

作　　者：尚海军 白新奎 乔磊 邓秦生 白旭 李恭斌 孙玉成 尹军波 刘圣冠 耿如意 Shang Hai-jun;Bai Xin-kui;Qiao Lei;Deng Qin-sheng;Bai Xu;Li Gong-bin;Sun Yu-cheng;Yin Jun-bo;Liu Sheng-guan;Geng Ru-yi(Xi'an West Heat Energy Saving Technology Co.,Ltd.;Hua'neng Gansu Energy Development Co.,Ltd.;Huaneng Lanzhou New District Thermal Power Co.,Ltd.)

机构地区：[1]西安西热节能技术有限公司 [2]华能甘肃能源开发有限公司 [3]华能兰州新区热电有限公司

出　　处：《建筑热能通风空调》2022年第9期6-8,21,共4页Building Energy & Environment

基　　金：中国华能集团有限公司总部科技项目(HNKJ21-HF258)。

摘　　要：提出一种基于深度置信网络的区域供热逐时负荷预测方法,并以兰州新区某换热站实际运行数据对所提出方法的有效性进行验证。此外,为分析建筑物热惰性对供热逐时负荷预测精确度的影响,分别将预测时刻前1 h,1~2 h,和1~3 h时作为输入参数的时间序列。研究结果表明:当时间序列取为预测时刻前1 h时显示出最佳的预测性能,预测值与实际值的平均绝对误差和平均相对误差分别为277.98 k W和2.28%,且相比采用人工神经网络分别降低约17.56 k W和0.15%。An hourly load prediction method for district heating based on Deep belief Network is proposed, and the effectiveness of the proposed method is verified by the actual operation data of a heat exchange station in Lanzhou. In addition, in order to analyze the influence of building thermal inertia on the accuracy of hourly heating load prediction,1 h, 1-2 h, and 1-3 h before the prediction time were taken as the time series of input parameters respectively. The results show that the optimal prediction performance is obtained when the time series is 1 h before the prediction time, and the average absolute error and average relative error between the predicted value and the actual value are 277.98 k W and2.28%, respectively, which are about 17.56 kW and 0.15% lower than the artificial neural network.

关 键 词：区域供热 供热负荷预测 深度置信网络 热惰性 

分 类 号：TU995[建筑科学—供热、供燃气、通风及空调工程] TP18[建筑科学—市政工程]