引用本文:蔡延华,张浩博,麦宇豪,等.基于数据驱动的可解释性低压台区理论线损率计算方法[J].电力系统保护与控制,2026,54(07):57-68.
CAI Yanhua,ZHANG Haobo,MAI Yuhao,et al.Data-driven interpretable method for calculating theoretical line loss rate in low-voltage distribution areas[J].Power System Protection and Control,2026,54(07):57-68
【打印本页】   【下载PDF全文】   【查看/发表评论】  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被：浏览 315次   下载 1122次 本文二维码信息
码上扫一扫！
分享到： 微信 更多
字体:加大+|默认|缩小-
基于数据驱动的可解释性低压台区理论线损率计算方法
蔡延华1，张浩博2，麦宇豪1，孟繁琪1，赵卓立2，彭显刚2
1.广东电网有限责任公司湛江供电局，广东 湛江 524005；2.广东工业大学自动化学院，广东 广州 510006
摘要:
针对现有基于物理模型的理论线损率计算方法效率低难以在线应用，而数据驱动方法可解释性不足、制约其工程推广应用等问题，提出一种基于数据驱动的可解释性低压台区理论线损率计算方法。首先，系统梳理了影响理论线损率物理机制的相关文献，筛选出影响理论线损率的关联特征。其次，构建以轻量级梯度提升机(light gradient boosting machine, LightGBM)为核心的计算模型，采用改进灰狼优化算法(improved grey wolf optimizer, IGWO)进行超参数寻优，以提升模型的计算精度。并引入SHAP(shapley additive explanations)方法量化各特征对计算值的贡献，揭示模型决策逻辑与理论线损率变化背后的物理损耗机理是否一致。最后，基于真实台区数据集的算例验证所提方法的有效性。
关键词:  可解释性  理论线损率计算  LightGBM  IGWO  SHAP
DOI：10.19783/j.cnki.pspc.250930
分类号:
基金项目:国家自然科学基金项目资助(62273104)；广东电网有限责任公司科技项目资助(GDKJXM20231545)
Data-driven interpretable method for calculating theoretical line loss rate in low-voltage distribution areas
CAI Yanhua1, ZHANG Haobo2, MAI Yuhao1, MENG Fanqi1, ZHAO Zhuoli2, PENG Xiangang2
1. CSG Guangdong Zhanjiang Power Supply Bureau, Zhanjiang 524005, China; 2. School of Automation, Guangdong University of Technology, Guangzhou 510006, China
Abstract:
To address the limitations of existing theoretical line loss rate calculation methods, namely, the low efficiency and difficulty of online application in physics-based models, and the lack of interpretability in data-driven approaches that hinders their engineering deployment, this paper proposes an data-driven interpretable method for calculating theoretical line loss rates in low-voltage distribution areas. First, relevant literature on the physical mechanisms affecting theoretical line loss rate is systematically reviewed, and key influencing features are identified. Second, a calculation model is constructed with the light gradient boosting machine (LightGBM) as its core, and an improved grey wolf optimizer (IGWO) is employed for hyperparameter optimization to enhance model accuracy. Furthermore, the shapley additive explanations (SHAP) method is introduced to quantify the contribution of each feature to the calculated results, thereby revealing whether the model’s decision logic aligns with the underlying physical mechanisms of line losses. Finally, case studies based on real-world distribution area datasets demonstrate the effectiveness of the proposed method.
Key words:  interpretability  theoretical line loss rate calculation  LightGBM  IGWO  SHAP
  • 1
X关闭
  • 1
X关闭