摘要
为更全面、准确地评估10 kV配电网线损水平,提出了一种基于灰色关联分析和改进神经网络的10kV配电网线线损预测方法。通过灰色关联分析方法定量分析了15个电气指标与10 kV配电网线损的关联性,再经过实际10 kV配电网数据的预测校验,最终确定了最佳的电气特征指标体系;其次使用十折交叉验证法结合试凑法计算分析不同神经网络结构下的模型预测性能,确定了最佳的网络结构,解决了BP神经网络(BPNN)隐含层节点数目多凭经验确定的缺点。考虑到传统的BP神经网络收敛速度慢、易陷入局部极小等缺点,采用自适应遗传算法改进BP神经网络(AGABPNN)的方法,进行学习和预测,并对比分析了该方法和径向基神经网络(RBFNN)、传统的BP神经网络的收敛性和预测准确性。通过某地区329条10kV线路实例计算,3种方法最小预测误差分别为6.71%、12.95%、17.05%,验证了AGA-BPNN具有更好的收敛性和泛化能力。
To estimate the level of 10 kV distribution network line loss more integrally and accurately, a 10 k V distribution network line loss prediction method based on grey correlation analysis and improved artificial neural network is proposed. Relevancies between 15 electrical indexes and 10 kV distribution network line loss are analyzed with grey correlation, and the best electrical characteristic indexes are selected after checked with practical data of 10 kV distribution network. To overcome difficulty in determining the number of nodes in hidden layer, predictive performances of the line loss prediction model under different neural network structures are analyzed with cross validation and test-and-error methods.Considering slow convergence and existence of local minimum in conventional BP neural network(BPNN), adaptive genetic algorithm(AGA) is adopted to improve BP neural network(AGA-BPNN), and compared with RBF neural network(RBFNN) and conventional BP neural network. After calculating actual data of 10 kV lines, minimum prediction error for 3 methods above are 6.71%, 12.95% and 17.05%respectively, proving better convergence and accuracy of AGA-BPNN.
作者
张义涛
王泽忠
刘丽平
邓春宇
孙云超
王新迎
韩笑
ZHANG Yitao;WANG Zezhong;LIU Liping;DENG Chunyu;SUN Yunchao;WANG Xinying;HAN Xiao(School of Electrical and Electronic Engineering,North China Electric Power University,Changping District,Beijing 102206,China;China Electric Power Research Institute,Haidian District,Beijing 100192,China)
出处
《电网技术》
EI
CSCD
北大核心
2019年第4期1404-1410,共7页
Power System Technology
基金
国家电网公司科技项目(面向同期线损管理的多专业数据治理技术与挖掘应用研究(XT71-17-027))~~
关键词
10kV配电网线损
灰色关联分析
神经网络
自适应遗传算法
10kV distribution network line loss
grey correlation analysis
artificial neural network
adaptive genetic algorithm
作者简介
通信作者:张义涛(1992),男,硕士研究生,研究方向为人工智能算法在配电网中的应用,E-mail:1714615688@qq.com;王泽忠(1960),男,教授,博士生导师,研究方向为电力系统电磁兼容和电磁场数值计算,E-mail:cememc@163.com;刘丽平(1964),女,教授级高级工程师,研究方向为电力系统及其自动化,E-mail:liulp@epri.sgcc.com.cn。
