深度学习在电力负荷预测中的应用综述  

Deep Learning Applications in Power System Load Forecasting: a Survey

作　　者：朱俊丞 杨之乐 郭媛君[2] 于坤杰 张建康[4] 穆晓敏[4] ZHU Juncheng;YANG Zhile;GUO Yuanjun;Yu Kunjie;ZHANG Jiangkang;MU xiaomin(Institute of industrial technology, Zhengzhou University, Zhengzhou 450001, China;Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518000, China;School of Electrical Engineering, Zhengzhou University, Zhengzhou, 450001, China;School of Information Engineering, Zhengzhou University, Zhengzhou, 450001, China)

机构地区：[1]郑州大学产业技术研究院,河南郑州450001 [2]中国科学院深圳先进技术研究院,广东深圳518000 [3]郑州大学电气工程学院,河南郑州450001 [4]郑州大学信息工程学院,河南郑州450001

出　　处：《郑州大学学报（工学版）》2019年第5期12-21,共10页Journal of Zhengzhou University（Engineering Science）

基　　金：国家自然科学基金资助项目(51607177、61876169、61806179、61433012、U1435215);广东省自然科学基金项目(2018A030310671);国家博士后科学基金面上项目(2018M631005)

摘　　要：在综合能源系统和能源互联网的高速发展中,电力负荷预测对电力系统的经济安全运行具有重要的作用.传统的负荷预测模型方法已在电力系统中取得了广泛应用,传统方法的简单计算模型对于高随机性、大数据背景下的动态负荷预测精度无法保证.近年来,在计算工具不断升级和训练数据量大规模提升的背景下,深度学习方法在电力负荷预测领域的应用得到了广泛重视.对多种深度学习方法在负荷预测领域中的应用进行了叙述分析,回顾了循环神经网络(RNN)、长短期记忆网络(LSTM)、深度置信网络(DBN)、卷积神经网络(CNN)等不同深度学习方法预测模型.对比于传统的负荷预测方法,深度学习方法具有更高的预测精度,对于各种外部影响因素具有更好的鲁棒性.In the rapid development of integrated energy systems and energy network, power load forecasting played an important role in the economic and safe operation of energy and power systems. The traditional load forecasting modelling methods have been widely used in power systems. However, the simple computational model structure limited by traditional methods could not guarantee the dynamic load prediction accuracy under high randomness and big data background. In recent years, in the context of the continuous upgrading of computing tools and the increasing large-scale of training data volume, the application of deep learning methods in the field of power system load forecasting atrracted extensive attentions. This paper analyzed the applications of various deep learning methods in the field of load forecasting, and revieed the Recurrent Neural Network (RNN), Long- and Short-Term Memory Network (LSTM), Deep Belief Network (DBN), and Convolutional Neural Network (CNN). Compared with the traditional load forecasting method, the deep learning method showed higher prediction accuracy and better robustness to various external influences.

关 键 词：深度学习 电力系统 负荷预测 人工神经网络 LSTM 

分 类 号：TU528.1[建筑科学—建筑技术科学]