Page 129 - 《中国电力》2026年第3期

第 59 卷 第 3 期 Vol. 59, No. 3
2026 年 3 月 ELECTRIC POWER Mar. 2026

引用格式：蒋达飞, 艾洪克, 孟巧, 等. 基于改进 ISODATA 算法的变电站负荷特性聚类[J]. 中国电力, 2026, 59(3): 125−133.
Citation: JIANG Dafei, AI Hongke, MENG Qiao, et al. Clustering of substation load characteristics based on improved ISODATA algorithm[J].
Electric Power, 2026, 59(3): 125−133.

基于改进 ISODATA 算法的变电站负荷特性聚类

蒋达飞 ，艾洪克 ，孟巧 ，董彪 ，翁一帆 ，张谦 2
2
1
1
1
1
(1. 国网冀北电力有限公司唐山供电公司，河北 唐山 063000；
2. 输变电装备技术全国重点实验室（重庆大学），重庆 400044)
Clustering of substation load characteristics based on
improved ISODATA algorithm

JIANG Dafei , AI Hongke , MENG Qiao , DONG Biao , WENG Yifan , ZHANG Qian 2
1
2
1
1
1
(1. Tangshan Power Supply Company of State Grid Jibei Electric Power Co., Tangshan 063000, China; 2. State Key Laboratory of Power
Transmission Equipment Technology (Chongqing University), Chongqing 400044, China)

Abstract: The new power system's high-voltage distribution dimensional space clustering, achieving explicit decoupling and
grid faces challenges posed by the large-scale and diversified clustering analysis of high-dimensional features. Simulation
connection of loads. Substation load clustering is a core method results indicate that in terms of feature extraction capability, the
for accurately identifying user electricity consumption patterns principal component analysis (PCA) feature space generated by
and optimizing grid resource allocation. Its analysis results can the improved algorithm exhibits significant differences in the
directly support grid planning, demand-side management, and seasonal load characteristics of substations, enabling better
the formulation of renewable energy integration strategies. capture of high-dimensional load features; In terms of algorithm
Therefore, it is urgent to conduct substation load curve performance, the improved algorithm reduces execution time by
clustering analysis to precisely analyze differentiated load 32.8%, lowers the Davies-Bouldin Index (DBI) by 29.1%, and
patterns and their dynamic evolution patterns, thereby providing increases the Dunn Index (DI) by 42.9%, validating the
data support for intelligent distribution grid operation decisions. effectiveness and superiority of the proposed algorithm.
Addressing the limitations of the iterative self-organizing data This work is supported by National Natural Science Foundation
analysis techniques algorithm (ISODATA), such as slow of China (No.52277081).
convergence speed and difficulty in capturing high-dimensional Keywords: substation; load clustering; clustering effect
data features—particularly the insufficient capture of load data's indicator
dynamic characteristics —this study enhances the algorithm's
摘 要：新型电力系统高压配电网面临规模化、多元化
ability to analyze high-dimensional features of substation load
负荷接入的挑战。变电站负荷聚类是精准识别用户用电
curves by optimizing the initial cluster center selection strategy
规律、优化电网资源配置的核心手段，其分析结果可直
and introducing a kernel function mapping mechanism. After
接 支 撑 电 网 规 划 、 需 求 侧 管 理 及 新 能 源 消 纳 策 略 制 定 。
completing missing value filling and data standardization
因此亟须通过变电站负荷曲线聚类分析，精准解析差异
preprocessing, this algorithm first optimizes the selection of 化负荷模式及其动态演化规律，为智能配电网运行决策
initial clustering centers based on the maximum distance 提供数据支撑。针对迭代式自组织数据分析算法（iterative
criterion to maximize the heterogeneity between initial centers self organizing data analysis techniques algorithm， ISODATA）
and improve clustering stability. Second, it introduces a kernel 存在收敛速度慢和难以捕捉数据高维特征的局限，尤其
function mapping mechanism to map load curves to high- 是负荷数据的动态特性捕捉不足的问题，分别通过优化
初始聚类中心选取策略与引入核函数映射机制，以提升
收稿日期：2025−05−28； 修回日期：2025−12−16。 算法对变电站负荷曲线高维特征的解析能力。在完成缺
基金项目：国家自然科学基金资助项目（52277081）。 失值填补与数据标准化预处理后，本算法首先基于最大

125