Scholarlink

<p>This paper presents a sensitivity analysis aimed at quantifying the impact of various features on the number of faults in medium voltage distribution networks. The analysis is based on the application of a neural network, trained on field data at disposal of the distribution network operator and pre-processed to obtain a numerical and balanced dataset. The analysis is conducted by perturbing the input features both one by one and in pairs, so to highlight the possible correlation between different features on fault occurrence. The dataset exploited for this paper regards the existing medium voltage distribution network of Rome, Italy, recorded from 2020 to 2023. This dataset collects records of several heterogeneous data sources referred to the normal operation (e.g., temperature and rainfall) and constituency (e.g., cable type, line length). Then, the dataset has been preprocessed since only a small percentage of the 22&#xa0;million records regards faults. The trained neural network and the related sensitivity analysis shows that the number of joints per branch and the ambient temperature have the main influence on fault occurrence. In particular, the greatest positive correlation with the number of predicted faults is found for the number of cable joints per branch. A 100% increase of the number of faults is predicted when the number joints per branch varies from 0 to 25.</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

Artificial Neural Network for Medium Voltage Fault Sensitivity Analysis

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Tommaso Bragatto,
  • Fabio Massimo Gatta,
  • Alberto Geri,
  • Marco Maccioni,
  • Alessandro Palazzoli,
  • Paolo Sancioni

摘要

This paper presents a sensitivity analysis aimed at quantifying the impact of various features on the number of faults in medium voltage distribution networks. The analysis is based on the application of a neural network, trained on field data at disposal of the distribution network operator and pre-processed to obtain a numerical and balanced dataset. The analysis is conducted by perturbing the input features both one by one and in pairs, so to highlight the possible correlation between different features on fault occurrence. The dataset exploited for this paper regards the existing medium voltage distribution network of Rome, Italy, recorded from 2020 to 2023. This dataset collects records of several heterogeneous data sources referred to the normal operation (e.g., temperature and rainfall) and constituency (e.g., cable type, line length). Then, the dataset has been preprocessed since only a small percentage of the 22 million records regards faults. The trained neural network and the related sensitivity analysis shows that the number of joints per branch and the ambient temperature have the main influence on fault occurrence. In particular, the greatest positive correlation with the number of predicted faults is found for the number of cable joints per branch. A 100% increase of the number of faults is predicted when the number joints per branch varies from 0 to 25.