基于ALIF-LSTM多任务学习的综合能源系统短期负荷预测  

SHORT-TERM LOAD FORECASTING METHOD FOR INTEGRATED ENERGY SYSTEM BASED ON ALIF-LSTM AND MULTI-TASK LEARNING

作　　者：欧阳静[1] 杨吕 尹康 赵宇航 潘国兵[1] Ouyang Jing;Yang Lyu;Yin Kang;Zhao Yuhang;Pan Guobing(Key Laboratory of E&M,Ministry of Education&Zhejiang Province,Zhejiang University of Technology,Hangzhou 310014,China;Zhejiang Huayun Electric Power Engineering Design Consulting Co.Ltd,Hangzhou 310014,China)

机构地区：[1]特种装备制造与先进加工技术教育部/浙江省重点实验室,浙江工业大学,杭州310014 [2]浙江华云电力工程设计咨询有限公司,杭州310014

出　　处：《太阳能学报》2022年第9期499-507,共9页Acta Energiae Solaris Sinica

基　　金：浙江省基础公益技术研究计划(LGF21E070001);国家重点研发计划(2017YFA0700300)。

摘　　要：综合能源系统中风电、光伏等可再生能源出力具有波动性和间歇性,精准的短期负荷预测有利于平抑可再生能源对系统运行的影响。系统中的多元负荷时间序列为典型的非平稳性信号,难以进行精准地预测。为了从数据层面提高综合能源系统短期负荷预测模型的精度,提出基于自适应局部迭代滤波(ALIF)的历史负荷数据分解方法,将历史负荷序列分解为具有不同频段模态函数的多个分量;针对预测模型训练中长时间序列处理困难及系统中多元负荷间耦合信息挖掘利用的问题,建立基于长短期记忆(LSTM)网络多任务学习的综合能源系统短期负荷预测模型。实验结果显示,与LSTM、ALIF-LSTM单任务学习、随机森林、LGBM方法相比,所提方法能够应对负荷波动剧烈的工况,预测精度较高,满足综合能源系统安全稳定运行控制的要求。The output of renewable sources such as wind power and photovoltaics in the integrated energy system is volatile and intermittent,and accurate short-term load forecasting is beneficial to smooth the impact of renewable energy on system operation. Shortterm load prediction is the basis for the safe and stable operation of the integrated energy system,while the time series of multiple loads of the system is a typical non-stationary signal with strong randomness,which makes it difficult to make an accurate prediction. In order to improve the accuracy of the short-term load prediction model of the integrated energy system from the data level,a historical load data decomposition method based on ALIF is proposed,which decomposes the historical load sequence into multiple components with different frequency band modal functions. In order to solve the problem of long time series processing in predictive model training and coupling information mining among multiple loads in the system,a short-term load forecasting model for integrated energy system based on LSTM multi-task learning was proposed. Experimental results show that,compared with LSTM,ALIF-LSTM single-task learning,random forest,and LGBM,the proposed method can cope with the condition of severe load fluctuation,and has a higher prediction accuracy,which meets the requirements of safe and stable operation and control of the integrated energy system.

关 键 词：可再生能源 长短期记忆网络 多任务学习 自适应局部迭代滤波 负荷预测 综合能源系统 

分 类 号：TK019[动力工程及工程热物理]