基于LSTM网络的居民区供热负荷短时预测  

Short-Term Prediction Based on LSTM Network for Heating Load of Residential Area

作　　者：许裕栗 姜娜 陈卓[3] 李柠[3] 甘中学 XU Yuli;JIANG Na;CHEN Zhuo;LINing;GAN Zhongxue(ENN Science and Technology Development Co.,Ltd.,Langfang 065001,China;State Key Laboratory of Coal-Based Low-Carbon Energy,Langfang 065001,China;Department of Automation,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]新奥科技发展有限公司,河北廊坊065001 [2]煤基低碳能源国家重点实验室,河北廊坊065001 [3]上海交通大学自动化系,上海200240

出　　处：《自动化仪表》2018年第10期1-5,10,共6页Process Automation Instrumentation

基　　金：国家重点基础研究发展(973)计划基金资助项目(2014CB249200)

摘　　要：用户侧供能负荷预测是能源互联网的关键技术之一,在能源互联网的运行管理中起着重要的作用。高精度的用户侧供能负荷预测能够提高能源利用率,有助于错峰资源的分层优化与自动分配,从而实现能源生产与使用的协同调度。现有的供能负荷预测方法很少考虑数据内部的时序相关性。基于实际的居民区历史供水热量数据,并考虑数据的时序特性,采用了长短时记忆(LSTM)网络,以挖掘居民区供水热量数据在时间维度上的内在联系。通过研究数据潜在的特征,建立了一种基于LSTM网络的居民区供热负荷短时预测模型。试验证明,相较于传统数据驱动的供热负荷短时预测模型,基于LSTM网络的居民区供热负荷短时预测模型具有更高的预测精度,在工业中具有更高的实际应用价值。未来将扩大数据集规模,引入与供热负荷相关的气象数据,构建基于LSTM网络的多步预测模型与供热需求长时预测模型。User-side energy load prediction is one of the key technologies of energy internet, and plays an important role in operation of energy intcrnct. High-accuracy user-side energy load prediction can improve energy utilization rate, and achieve the hierarchical optimization and automatic allocation of peak resources, so as to fulfill the collaborative scheduling of energy production and usage. In existing energy load forecasting methods,the temporal correlation in data is seldomly taken into account. Based on the historical data of residential hot water supply, and considering the temporal characteristics of the data, the long short-teml memory （LSTM） network is used to discover the internal temporal relations of residential hot water supply data. By studying the potential characteristics of the data, a residential heating load short term prediction model based on LSTM network is proposed. The test results show that the residential heating load short term prediction-model based on LSTM network has a hgher prediction accuracy than that of the traditional heating load short-time prediction models. The approach has high application values in the industry. In the future, the dataset will be expanded through adding meteorological data related to the heating load, and a multi-step prediction model based on LSTM network will be built for a long-ternl prediction of heating demand.

关 键 词：能源互联网 人工智能 供热负荷 短时预测 长短时记忆 循环神经网络 

分 类 号：TH183[机械工程—机械制造及自动化] TP29[自动化与计算机技术—检测技术与自动化装置]