基于神经网络的区域供热系统短期负荷预测  

Short-term loadprediction of district heating system based on Neural Network

作　　者：韩子俊 陈世新 申娜 田永红 任效效 冯朴 王进仕 HAN Zijun;CHEN Shixin;SHEN Na;TIAN Yonghong;REN Xiaoxiao;FENG Pu;WANG Jinshi(China Energy Construction Group Shaanxi Electric Power Design Institute Co.LTD,X'i an 710043,China;Xi'an Jiaotong University,School of Energy and Power Engineering,Xi'an 710049,China;China Huadian Power Generation Operation Co.LTD,Beijing 100031,China)

机构地区：[1]中国能源建设集团陕西省电力设计院有限公司,陕西西安710043 [2]西安交通大学能动学院,陕西西安710049 [3]中国华电集团发电运营有限公司,北京100031

出　　处：《区域供热》2023年第1期144-151,共8页District Heating

基　　金：陕西省创新能力支撑计划项目(2018TD-014)。

摘　　要：准确的热负荷预测是实现区域供热系统精细控制和节能减碳的关键。以国内北方某城市区域供热系统为研究对象,分别采用BP神经网络(BPNN)、遗传算法(Genetic Algorithm,GA)优化BP神经网络(GA-BPNN)和自回归移动平均模型(Autoregressive Integrated Moving Average Model,ARIMA)组合BP神经网络(ARIMA-BPNN)方法对其热负荷进行预测,并对比了各预测方法的准确性和适用性。结果表明,GA-BPNN预测误差最小,ARIMA-BPNN次之,但后者预测所需数据更少。此外,验证了在减少样本数目以及影响因素的种类的情况下,GA-BPNN预测方法的平均相对误差均在5%以内,表明GA-BPNN预测方法适用于样本减少的情况。Accurate prediction of heat load is the key to achieve fine control,save energy and reduce emissions of district heating system.This paper takes a district heating system in northern China as the research object,and BP neural network,genetic algorithm optimized BP neural network(GA-BPNN)and autoregressive integrated moving average model combined BP neural network(ARIMA-BPNN)are used to predict the heat load,and the accuracy and applicability of each prediction method are compared.The results show that the GA-BPNN has the smallest prediction error and the ARIMA-BPNN is the second,but the latter requires less data for prediction.In addition,it is verified that the average relative error of the GA-BPNN is less than 5%when the number of samples and the types of influencing factors are reduced,which indicates the GA-BPNN is suitable for the case of sample reduction.

关 键 词：BP神经网络 遗传算法 差分自回归移动平均模型 负荷预测 区域供热 

分 类 号：TU995[建筑科学—供热、供燃气、通风及空调工程] TP183[建筑科学—市政工程]