融合负荷引力聚类与深度Q网络的智能化台区设计  

Intelligent transformer area design combining load gravity clustering and deep Q-network

作　　者：奚振乾 韩祝明 王亚辉 任新丁 崔秋实 XI Zhenqian;HAN Zhuming;WANG Yahui;REN Xinding;CUI Qiushi(State Grid Anhui Electric Power Co.,Ltd.,Hefei 230022,China;Anhui Mingsheng Heng zhuo Technology Co.,Ltd.,Hefei 230031,China;School of Electrical Engineering,Shanghai University of Electric Power,Shanghai 200090,China;Dafang Software Co.,Ltd.,Zhengzhou 450052,China;School of Electrical Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]国网安徽省电力有限公司,安徽合肥230022 [2]安徽明生恒卓科技有限公司,安徽合肥230031 [3]上海电力大学电气工程学院,上海200090 [4]郑州大方软件股份有限公司,河南郑州450052 [5]重庆大学电气工程学院,重庆400044

出　　处：《电工电能新技术》2024年第8期105-112,共8页Advanced Technology of Electrical Engineering and Energy

摘　　要：随着电网的深化改革,对电网企业的精准投资和合理运行也提出了更高的要求。传统的配电台区规划设计,缺少科学的计算模型,导致设计阶段即出现供电距离长、导线曲折系数大等一系列问题。为了更好地推进配电台区规划设计方案科学制定,本文提出了一种基于强化学习的台区智能一体化规划设计方法。本文首先利用用户的历史负荷数据,采用负荷间引力的密度聚类算法对用户区域进行合理地划分,确定负荷中心。在划分好的区域内,采用深度强化学习算法对用户点、负荷中心点以及障碍物进行建模,设定最优路线的损失函数,对乡镇台区低压线路拓扑结构进行合理的规划设计。最后基于规划设计的拓扑及用户历史用电数据开展节点中性线电流最小计算,给出单相用户到相计算,实现低压电网中性线损耗最低。本文利用安徽省某地区的实际情况进行验证,使用本文所提的方法改造低压台区,变压器的布局偏移负荷中心点仅17 m,台区的线损率为1.5%,低于2.0%。With the deepening reform of the power grid,higher requirements have also been put forward for the precise investment and rational operation of power grid enterprises.The traditional planning and design of substation area lacks a scientific calculation model,which leads to a series of problems such as long power supply distance and large wire tortuosity coefficient in the design stage.In order to better promote the scientific formulation of the planning and design scheme of the distribution station area,this paper proposes an intelligent integrated planning and design method for the station area based on reinforcement learning.The paper firstly uses the user’s historical load data,adopts the density clustering algorithm considering the gravitational force between loads to divide the user area reasonably,and determines the load center.In the divided area,deep reinforcement learning algorithm is used to model the user points,load center points and obstacles,and the loss function of the optimal route is set,and the topology of the low-voltage line in the township transformer area is reasonably planned and designed.Finally,based on the topology of the planning and design and the user’s historical power consumption data,the minimum calculation of the node neutral current is carried out,and the single-phase user-to-phase calculation is given to achieve the lowest neutral loss of the low-voltage power grid.We used the actual situation of a region in Anhui province for verification,and used the method proposed in this paper to transform the low-voltage platform area.The transformer layout offset load center point was only 17 m,and the line loss rate of the platform area was decreased to 1.5%,lower than 2.0%.

关 键 词：台区智能规划设计 负荷引力 密度聚类 深度强化学习 

分 类 号：TM76[电气工程—电力系统及自动化]