基于Transformer算法的园区综合能源需求预测  

Integrated energy demand forecasting for the park based on the Transformer algorithm

作　　者：尹宇晨 刘宇杭 马愿谦 雷一 YIN Yuchen;LIU Yuhang;MA Yuanqian;LEI Yi(School of Information Science and Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China;Tsinghua Sichuan Energy Internet Research Institute,Chengdu 610200,China)

机构地区：[1]浙江理工大学信息科学与工程学院,杭州310018 [2]清华四川能源互联网研究院,成都610200

出　　处：《综合智慧能源》2023年第10期61-69,共9页Integrated Intelligent Energy

基　　金：浙江省自然科学基金项目(LQ22E070009);浙江理工大学科研业务费专项资金资助项目(23222130-Y)。

摘　　要：准确的综合能源需求预测是区域综合能源系统调度和能效评估的基础。在综合能源需求预测方面,影响因素众多、设计参数复杂、计算效率较低,且在长序列预测上仍有较大优化空间,因此提出了一种基于Transformer算法的园区综合能源需求预测方法。建立了园区冷热电负荷影响因素的筛选模型,为数据预处理后筛选适当的影响因素提供基础;建立了基于欧氏距离的综合相似度的相似日选取方法,为综合能源的预测奠定了基础;建立了基于Transformer算法的冷热电负荷预测模型,以实现园区综合能源需求预测;以中国东部某园区为对象进行算例分析,预测园区综合能源需求。结果表明,所提预测方法能有效提高预测精度,具有较高的准确度和实用性。Accurate forecasting on integrated energy demands will be the basis for the scheduling and energy efficiency assessment of regional integrated energy systems.Integrated energy demand forecasting is infected by multiple factors.And being hampered by complex design parameters and low calculation efficiency,there is plenty of room for the optimization of current long series prediction method.Therefore,an integrated energy demand forecasting method based on Transformer algorithm is proposed.Firstly,influencing factors are screened from pre-processed data by the influence factor selection model for the cooling,heating and electricity loads in a park.Secondly,similar days are categorized based on the Euclidean distance,which lays a foundation for the integrated energy prediction.Then,a forecasting model for cooling,heating and electricity loads based on Transformer algorithm is established to predict the integrated energy demand in the park.Finally,the proposed forecasting model is tested on a park located in eastern China,and the results verified its prediction accuracy and effectiveness.

关 键 词：园区综合能源 Transformer算法 能源需求预测 相似日 冷热电负荷 

分 类 号：TK01[动力工程及工程热物理]