基于神经网络的区域综合能源系统多时间尺度功率预测  

Multi-time scale power prediction of RIES based on neural network

作　　者：王炼 窦真兰 施超寅 夏景 徐瀚诚 贺林钰 WANG Lian;DOU Zhenlan;SHI Chaoyin;XIA Jing;XU Hancheng;HE Linyu(Songjiang Power Supply Company,State Grid Shanghai Electric Power Company,Shanghai 201600,China;State Grid Shanghai Integrated Energy Service Co.,Ltd.,Shanghai 200023,China)

机构地区：[1]国网上海市电力公司松江供电公司,上海201600 [2]国网上海综合能源服务有限公司,上海200023

出　　处：《电气应用》2023年第2期40-48,共9页Electrotechnical Application

摘　　要：考虑到区域综合能源系统(RIES)中可再生能源及负荷需求的时变性、不确定性等特点,提出基于长短期记忆神经网络和卷积神经网络的短期-超短期多时间尺度功率预测方法,为区域综合能源系统的优化调度提供有效准确的数据支撑。首先,应用广泛处理较长时间序列问题的长短期记忆神经网络(LSTM),在采用数据处理以及贝叶斯优化求最佳超参数的基础上,实现区域综合能源系统供需侧日前短期功率预测。为进一步提高预测准确度,利用卷积神经网络(CNN)在数据提取方面的高效性,同时采用双向长短期记忆神经网络(BiLSTM)进行双向训练,实现RIES供需侧日内超短期功率预测。最后,根据评价指标通过仿真验证了所提预测方法在RIES日前-日内优化调度中的有效性和适用性。Considering the time-varying and uncertain characteristics of renewable energy and load demand in the regional integrated energy system, a multi-time scale power prediction method based on long short-term memory neural network and convolutional neural network is proposed, which will provide effective and accurate data support for the optimal dispatch of regional integrated energy system. First, the short-term power prediction of the RIES supply and demand side based on data processing and Bayesian optimization to find the best hyperparameters is realized by using the LSTM neural network which widely deals with long-time series problems. To further improve the prediction accuracy, the intraday ultra-short-term power prediction on the supply and demand side of RIES is realized using the efficiency of CNN in data extraction and the bidirectional training of BiLSTM. Finally,the effectiveness and applicability of the proposed forecasting method in RIES multi-time scale optimal scheduling are verified by simulations according to the evaluation index.

关 键 词：神经网络 多时间尺度 功率预测 区域综合能源系统 评价指标 

分 类 号：TK01[动力工程及工程热物理] TM73[电气工程—电力系统及自动化]