Scholarlink

<p>This study presents a two-stage distributed learning method with residual correction to improve the accuracy of load forecasting in power distribution feeders. Unlike conventional approaches that separately forecast either total feeder load or individual sectional loads, the proposed method captures their hierarchical and interdependent structure. In the first stage, sectional loads are forecast using voltage, current, and electrical load as input features, enabling physically meaningful modeling. These forecasted sectional loads are then aggregated to estimate the total feeder load. In the second stage, the residual between the estimated and actual total load is learned and corrected through an additional MLP model. This hybrid approach integrates physics-informed forecasting and data-driven refinement, offering enhanced accuracy, interpretability, and modularity. Experimental results show high forecasting performance with an NRMSE of 2.97%, a MAPE of 2.84%, and an R² score of 0.9931. The model effectively captures load dynamics even in highly variable sections, demonstrating its potential as a reliable tool for practical distribution system operations.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Study on the Load Forecasting of Distribution Feeder Based on Residual Correction Distributed Learning Considering Sectional Loads

  • 期刊论文

  • 发表日期：

    2026年1月2日

查看全文
  • Jun-Hyeok Kim

摘要

This study presents a two-stage distributed learning method with residual correction to improve the accuracy of load forecasting in power distribution feeders. Unlike conventional approaches that separately forecast either total feeder load or individual sectional loads, the proposed method captures their hierarchical and interdependent structure. In the first stage, sectional loads are forecast using voltage, current, and electrical load as input features, enabling physically meaningful modeling. These forecasted sectional loads are then aggregated to estimate the total feeder load. In the second stage, the residual between the estimated and actual total load is learned and corrected through an additional MLP model. This hybrid approach integrates physics-informed forecasting and data-driven refinement, offering enhanced accuracy, interpretability, and modularity. Experimental results show high forecasting performance with an NRMSE of 2.97%, a MAPE of 2.84%, and an R² score of 0.9931. The model effectively captures load dynamics even in highly variable sections, demonstrating its potential as a reliable tool for practical distribution system operations.