Reproducibility Package for "A Two-Branch TCN with Underforecast-Aware Correction for Short-Term Load Forecasting During Critical Periods"

This dataset provides the reproducibility package for the study "A Two-Branch TCN with Underforecast-Aware Correction for Short-Term Load Forecasting During Critical Periods". The package contains the implementation scripts, experimental configurations, processed data, and generated results required to reproduce the reported short-term load forecasting experiments. The study investigates critical-period underforecast risk in power-system load forecasting and proposes a two-branch temporal convolutional network framework combining a standard TCN-Huber branch and an underforecast-aware TCN-H3-CRW branch. The package includes: (1) Python scripts for frozen final validation and baseline audit on the Panama dataset; (2) the processed feature dataset used in the experiments; (3) generated experimental results and manuscript-related tables. The provided files correspond to the final experimental configuration used in the manuscript.本数据集为研究《面向关键时段短期负荷预测的带欠预测感知校正的双分支时间卷积神经网络（TCN，Temporal Convolutional Network）》提供了可复现数据包。 本数据包包含复现该研究报道的短期负荷预测实验所需的实现脚本、实验配置文件、预处理后数据集以及生成的实验结果。 该研究聚焦电力系统负荷预测中的关键时段欠预测风险，提出了一种双分支时间卷积神经网络（TCN）框架，该框架结合了标准TCN-Huber分支与欠预测感知TCN-H3-CRW分支。 本数据包包含以下内容： (1) 用于巴拿马数据集上的固定参数最终验证与基线模型审计的Python脚本； (2) 实验中使用的预处理特征数据集； (3) 生成的实验结果与论文相关表格。 本次提供的文件对应论文中使用的最终实验配置。