Solar Irradiation Forecasting Dataset (SURFRAD, 2018–2020): Reproducible Inputs for GA-Optimized Machine Learning Models

This repository contains the solar irradiation and meteorological data derived from the U.S. SURFRAD network for the years 2018–2020. The dataset supports the study “Solar Irradiation Forecasting Using Genetic Algorithms: GA-Optimized Machine Learning Models for Short-Term Solar Irradiation Prediction” and is designed to enable full reproducibility of all results reported in the manuscript. To reduce storage footprint and improve reproducibility, quality-control (QC) flags, redundant time fields, and unused meteorological variables were removed prior to analysis. Only variables explicitly used in feature selection, modeling, tables, and figures were retained. The dataset includes inputs required for Linear Regression (LR), Extreme Gradient Boosting (XGB), and Genetic Algorithm–optimized XGB models, with a strict temporal split: – Training data: 2018–2019 – Validation data: 2020 All data are observational, non-proprietary, and intended solely for methodological demonstration, benchmarking, and reproducibility.本仓库收录了2018至2020年源自美国SURFRAD网络的太阳辐射与气象数据。 本数据集配套支撑学术论文《基于遗传算法的太阳辐射预报：面向短期太阳辐射预测的遗传算法优化机器学习模型》，旨在实现该论文中所有报道结果的完全可复现。 为缩减存储占用并提升可复现性，研究人员在分析前已移除质量控制（Quality Control，QC）标记、冗余时间字段及未使用的气象变量，仅保留特征选择、建模、制表及绘图环节明确用到的变量。 本数据集包含线性回归（Linear Regression，LR）、极限梯度提升（Extreme Gradient Boosting，XGB）以及遗传算法（Genetic Algorithm，GA）优化XGB模型所需的输入数据，并采用严格的时间划分规则： —— 训练集：2018至2019年 —— 验证集：2020年 所有数据均为实地观测所得，无专有版权约束，仅用于方法演示、基准测试与结果可复现性验证。