基于K-Means聚类和改进多分类相关向量机的台区线损计算方法

doi:10.11985/2021.01.009

摘要/Abstract

摘要：

配电网线损计算是配电网线损管理和分析的一项重要技术措施。针对传统配电网理论计算不能自动实现、工作量大、计算结果不准确等问题,提出了一种基于K-Means聚类和果蝇算法优化多分类相关向量机(Multi classification correlation vector machine,MRVM)的快速计算低压台区线损计算方法,并在Matlab中搭建算法模型。针对台区线损数值分散的问题,先使用K-Means聚类算法将台区样本进行归类。使用归类的样本对果蝇算法优化的MRVM进行训练,得到线损率和台区参数之间的映射关系。以四川某地的500个台区样本进行实例验证,并与传统理论线损计算方法结果进行了对比。结果表明,所提出的方法可以有更好的计算精度。同时可以提高当前线损理论计算的自动化程度,减少计算工作量,提高电网公司配电网运行效率。

关键词: 配电网, 线损, K-Means聚类算法, 果蝇算法, 多分类相关向量机

Abstract:

The calculation of distribution network line loss is an important technical measure for the management and analysis of distribution network line loss. In order to solve the problems of traditional distribution network theory, such as unable to realize automatically, heavy workload and inaccurate calculation results, a fast calculation method of line loss in low-voltage substation area based on K-Means clustering and Drosophila algorithm optimization multi classification correlation vector machine (MRVM) is proposed, and an algorithm model is built in Matlab. In order to solve the problem of dispersion of line loss, K-Means clustering algorithm is used to classify the samples. Then, the MRVM optimized by the Drosophila algorithm is trained by using the classified samples, and the mapping relationship between the line loss rate and the platform area parameters is obtained. Finally, 500 samples of a certain area in Sichuan Province are used to verify the results, and the results are compared with the traditional theoretical line loss calculation method. The results show that the proposed method has better accuracy. At the same time, it can improve the automation degree of the current theoretical calculation of line loss, reduce the calculation workload, and improve the operation efficiency of the distribution network of the grid company.

Key words: Distribution network, line loss, K-Means clustering algorithm, Drosophila algorithm, MRVM

中图分类号:

TM744

谢林, 李红伟, 袁岳, 周海林. 基于K-Means聚类和改进多分类相关向量机的台区线损计算方法[J]. 电气工程学报, 2021, 16(1): 62-69.

XIE Lin, LI Hongwei, YUAN Yue, ZHOU Hailin. Calculation of Line Loss in Transformer District Based on K-Means Clustering Algorithm and Improved MRVM[J]. Journal of Electrical Engineering, 2021, 16(1): 62-69.

图/表 15

图1

图2

图3

图4

表1

表2

表3

图5

图6

图7

图8

图9

图10

表4

表5

参考文献 20

[1]	章元德, 史亮, 陆巍, 等. 线损信息化统计中数据质量管控机制及实现[J]. 电力系统自动化, 2016,40(7):128-133.
	ZHANG Yuande, SHI Liang, LU Wei, et al. Realization of data management and control mechanism of line loss statistical inofrmation[J]. Automation of Electric Power System, 2016,40(7):128-133.
[2]	高卫东, 宋斌. 月度实际线损率定量计算方法[J]. 电力系统自动化, 2012,36(2):86-90.
	GAO Weidong, SONG Bin. A quantitative calculation method for actual monthly line loss rate[J]. Automation of Electric Power System, 2012,36(2):86-90.
[3]	罗洁青. 广州荔湾区配电网线损分析及降损措施研究[D]. 广州:华南理工大学, 2012.
	LUO Jieqing. Studies on line loses of distribution networks and measures for reducing line loss in Liwan district of Guangzhou[D]. Guangzhou:South China University of Technology, 2012.
[4]	颜伟, 吕志盛, 李佐君, 等. 输电网的蒙特卡罗模拟与线损概率评估[J]. 中国电机工程学报, 2007(34):41-47.
	YAN Wei, LÜ Zhisheng, LI Zuojun, et al. Monte Carlo simulation and transmission loss evalution with probabilistic method[J]. Proceedings of the CSEE, 2007(34):41-47.
[5]	王守相, 周凯, 苏运. 基于随机森林算法的台区合理线损率估计方法[J]. 电力自动化设备, 2017,37(11):39-45.
	WANG Shouxiang, ZHOU Kai, SU Yun. Line loss rate estimation method of transformer districtbased on random forest algorithm[J]. Automation of Electric Power Systems, 2017,37(11):39-45.
[6]	邹云峰, 梅飞, 李悦, 等. 基于数据挖掘技术的台区合理线损预测模型研究[J]. 电力需求侧管理, 2015(4):25-29.
	ZOU Yunfeng, MEI Fei, LI Yue, et al. Prediction model research of reasonable line loss for transformer district based on data mining technology[J]. Power Demand Side Management, 2015(4):25-29.
[7]	姜惠兰, 安敏, 刘晓津, 等. 基于动态聚类算法径向基函数网络的配电网线损计算[J]. 中国电机工程学报, 2005(10):35-39.
	JIANG Huilan, AN Min, LIU Xiaojin, et al. The calculation of energy losses in distribution systems based on RBF network with dynamic clustering algorithm[J]. Proceedings of the CSEE, 2005(10):35-39.
[8]	李滨, 严康, 罗发, 等. 最优标杆在市级电网企业线损精益管理中的综合应用[J]. 电力系统自动化, 2018,42(23):184-192.
	LI Bin, YAN Kang, LUO Fa, et al. Comprehensive application of optimal benchmarking in line loss lean management of city-level power grid enteprises[J]. Automation of Electric Power Systems, 2018,42(23):184-192.
[9]	王潇笛, 刘俊勇, 刘友波. 采用自适应分段聚合近似的典型负荷曲线形态聚类算法[J]. 电力系统自动化, 2019,43(1):150-161.
	WANG Xiaodi, LIU Junyong, LIU Youbo. Shape clustering algorithm of typical load curves based on adaptive piecewise aggregate approximation[J]. Automation of Electric Power Systems, 2019,43(1):150-161.
[10]	张绍德, 毛雪菲, 毛雪芹. 基于最近邻聚类支持向量机辨识的电弧炉电极逆控制[J]. 控制理论与应用, 2010(7):85-91.
	ZHANG Shaode, MAO Xuefei, MAO Xueqin. Inverse control for electrodes in electric arc furnace based onsupport-vector-machines identification on nearest neighbor clustering[J]. Control Theory & Applications, 2010(7):85-91.
[11]	王保义, 胡恒, 张少敏. 差分隐私保护下面向海量用户的用电数据聚类分析[J]. 电力系统自动化, 2018,42(2):121-127.
	WANG Baoyi, HU Heng, ZHANG Shaomin. Clustering analysis of power consumption data to mass users under differential privacy protection[J]. Automation of Electric Power Systems, 2018,42(2):121-127.
[12]	CHITSAZ H, AMJADY N, ZAREIPOUR H. Wind power forecast using wavelet neural network trained by improved Clonal selection algorithm[J]. Energy Conversion & Management, 2015,89:588-598.
[13]	律方成, 金虎, 王子建, 等. 基于主成分分析和多分类相关向量机的GIS局部放电模式识别[J]. 电工技术学报, 2015,30(6):225-231.
	LÜ Fangcheng, JIN Hu, WANG Zijian, et al. GIS partial discharge pattern recognition based on principal component analysis and milticlass relevance vector machine[J]. Transactions of China Electrotechnical Society, 2015,30(6):225-231.
[14]	尹金良. 基于相关向量机的油浸式电力变压器故障诊断方法研究[D]. 北京:华北电力大学, 2013.
	YIN Jinliang. Fault diagnosis method of oil immersed power transformer based on relevance vector machine[D]. Beijing:North China Electric Power University, 2013.
[15]	吴小文, 李擎. 果蝇算法和5种群智能算法的寻优性能研究[J]. 火力与指挥控制, 2013(4):21-24,29.
	WU Xiaowen, LI Qing. Research of optimizing performance of fruit fly optimizationalgorithm and five kinds of intelligent algorithm[J]. Fire Control＆Command Control, 2013(4):21-24,29.
[16]	胡能发. 演化式果蝇算法及其应用研究[J]. 计算机技术与发展, 2013,23(7):131-133,137.
	HU Nengfa. Evolutionary fruit algorithm and its application research[J]. Computer Technology and Development , 2013,23(7):131-133,137.
[17]	周建英, 宋奕冰, 刘宪林, 等. 网络最大流法在配电网线损计算中的应用[J]. 电力系统及其自动化学报, 2000(5):13-16.
	ZHOU Jianying, SONG Yibing, LIU Xianlin, et al. Application of network maximal flow in the computation of distribution network losses[J]. Proceedings of the EPSA, 2000(5):13-16.
[18]	RAUTIO J C, DEMIR V. Microstrip conductor loss models for electromagnetic analysis[J]. IEEE Trans. Mtt., 2003,51(3):915-921.
[19]	姜惠兰, 安敏, 刘晓津, 等. 基于动态聚类算法径向基函数网络的配电网线损计算[J]. 中国电机工程学报, 2005(10):37-41.
	JIANG Huilan, AN Min, LIU Xiaojin, et al. The calculation of energy losses in distribution systems based on RBF network with dynamic clustering algorithm[J]. Proceedings of the CSEE, 2005(10):37-41.
[20]	徐茹枝, 王宇飞. 粒子群优化的支持向量回归机计算配电网理论线损方法[J]. 电力自动化设备, 2012,32(5):86-89.
	XU Ruzhi, WANG Yufei. Theoretical line loss calculation based on SVR and PSO for distribution system[J]. Electic Power Automation Equipment, 2012,32(5):86-89.

类别	初始聚类中心				P_E
类别	x₁(%)	x₂/m	x₃/m	x₄(%)	P_E
1	82.31	132	789	6.134	2.156
2	80.55	201	1 532	8.170	1.934
3	91.01	189	1 980	7.460	3.106
4	90.32	390	1 756	15.450	3.598
5	97.22	267	1 078	7.310	3.709
6	79.98	600	4 037	10.890	4.623

参数	训练目标误差(%)
参数	0.01	0.005	0.001	0.000 1
迭代次数	70	98	160	480
均方误差	1.50	0.67	0.17	0.09
E_C小于0.05%台数	0	7	9	39
E_C介于0.05%和0.1%之间台数	2	3	8	30
E_C介于0.1%和1%之间台数	38	40	141	317
E_C介于1%和5%台数	109	169	250	173
E_C介于5%和10%台数	108	119	80	19
E_C大于10%台数	343	262	112	22

样本号	线损特征指标				实测线损(%)	相对误差(%)
样本号	X₁(%)	X₂/m	X₃/m	X₄(%)	实测线损(%)	相对误差(%)
一类30号	98.1	246	490	11.33	11.26	25.7
四类5号	100	250	500	13.14	2.03	11.7
五类5号	85.2	200	470	9.1	15.03	21.4
五类15号	90.4	231	494	10.7	10.08	13.2
五类64号	101	236	500	12.1	7.09	17.6

参数	训练目标误差设为0.01
	第1类		第2类		第3类		第4类		第5类		第6类
	标准MRVM	改进MRVM	标准MRVM	改进MRVM	标准MRVM	改进MRVM	标准MRVM	改进MRVM	标准MRVM	改进MRVM	标准MRVM	改进MRVM
迭代次数	60 000	12	68 000	12	2 900	8	2 400	5	11 000	6	790	21
均方误差E_m(%)	41.3	2.1	28.4	4.1	33.8	2	24.4	3.2	16.7	2.4	31.1	1.8