"Operational Data Set, Cleaning, and Evaluation of AI Models for Predicting Faults in Medium-Voltage Reclosers"

"This study presents a time-series dataset collected from nine SEL three-phase reclosers installed in medium-voltage distribution networks of a utility in the Puno region, Peru, between January 2024 and June 2025. The dataset integrates two types of records: load measurements with a 15-minute resolution (156,023 observations) and protection system events (1,191 three-phase and single-phase faults). The electrical variables include phase and line-to-line voltages, currents, active and reactive power, power factor, frequency, accumulated energy, and sequence components.Data curation involved correcting date format inconsistencies, noise filtering, and temporal synchronization to ensure integrity and continuity. The dataset\u2019s uniqueness lies in its origin: real measurements from a Latin American utility, a region underrepresented in open power system datasets, and the careful selection of nine authorized reclosers from over 90 devices, based on data availability.This dataset enables the development and evaluation of machine learning methods for fault prediction, predictive maintenance, and time-series analysis in distribution networks, providing a valuable resource for both the data science and power systems communities."

本研究构建了一套时间序列数据集，采集自秘鲁普诺地区某公用电力公司中压配电网中部署的9台SEL三相重合器，采集时段为2024年1月至2025年6月。该数据集包含两类记录：分辨率为15分钟的负荷测量数据（共156023条观测记录），以及保护系统事件数据（包含1191起三相及单相故障事件）。所涵盖的电气变量包括相电压、线电压、电流、有功功率、无功功率、功率因数、频率、累计电能以及序分量。数据集经过了数据规整处理，包括修正日期格式不一致问题、噪声过滤以及时间同步操作，以确保数据的完整性与连续性。该数据集的独特性源于其采集来源：取自拉丁美洲某公用电力公司的实测数据，而该区域在开放电力系统数据集中的代表性不足；同时，研究团队基于数据可用性，从90余台设备中精心筛选出9台经授权的重合器。本数据集可用于开发并评估配电网场景下的故障预测、预测性维护以及时间序列分析相关的机器学习方法，可为数据科学与电力系统领域的科研人员提供极具价值的研究资源。