摘要
针对含风电光伏电力系统网损率计算的问题,提出一种基于深度迁移学习(transfer-deep boltzmann network-deep neural network,TDBN-DNN)的网损率计算模型。首先将训练好的深度学习模型作为源模型,并冻结深度置信网络(deep boltzmann network,DBN)特征提取层。然后定义了最大均值差异贡献数ρi,迁移出与待计算数据分布更接近的样本数据,微调深层神经网络(deep neural network,DNN),得到基于TDBN-DNN的网损率计算模型。最后以中国北方某地区实际电网为算例进行验证,仿真结果表明,该DBN-DNN深度学习计算方法较传统浅层结构的BP(back propagation)神经网络计算方法拥有更好的非线性拟合能力。此外,经过迁移学习后得到的深度迁移学习TDBN-DNN模型拥有更高的计算精度与更好的时效性,而且该模型具有一定的数据容错性。
Aiming at the problem of network loss calculation of wind power photovoltaic power system,a calculation model of network loss rate based on transfer-deep boltzmann network-deep neural network(TDBN-DNN)was proposed.Firstly,the trained deep learning model was used as the source model,and the deep boltzmann network(DBN)feature extraction layer was frozen.Then,contribution coefficient of maximum mean discrepancy was defined and the sample data that has closer distribution with the data to be calculated is selected.The sample data was used to fine-tune the deep neural network(DNN),and the network loss calculation model based on TDBN-DNN was obtained.The TDBN-DNN based deep migration learning model was used to calculate the network loss rate.Finally,the actual power grid in a certain area of northern China was taken as an example.The simulation results show that the DBN-DNN deep learning calculation method has better nonlinear fitting ability than the traditional shallow back propagation(BP)neural network calculation method.In addition,after migration learning,the TDBN-DNN deep migration learning model has higher calculation accuracy and better timeliness than the DBN-DNN model and also has certain data fault tolerance.
作者
卢志刚
杨英杰
李学平
陈建华
刘建恒
LU Zhigang;YANG Yingjie;LI Xueping;CHEN Jianhua;LIU Jianheng(Key Lab of Power Electronics for Energy Conservation and Motor Drive of Hebei Province,Yanshan University,Qinhuangdao 066004,Hebei Province,China;State Grid Jibei Power Co.,Ltd.,Xicheng District,Beijing 100054,China)
出处
《中国电机工程学报》
EI
CSCD
北大核心
2020年第13期4102-4110,共9页
Proceedings of the CSEE
基金
国家自然科学基金项目(61873225,61374098)。
关键词
深度学习
迁移学习
最大均值差异
网损率计算
deep learning
migration learning
maximum mean difference
network loss rate calculation
