QuaLiKiz-v2.6.2 turbulent transport model evaluations based on JET experimental plasma profiles

This dataset was used to train the QuaLiKiz-neural-network (QLKNN) model, QLKNN-jetexp-15D, described within the following published article: https://doi.org/10.1063/5.0038290. It was generated with approximately 33 million standalone evaluations of QuaLiKiz-v2.6.2, each performed with a standard vector of 18 wavenumbers. Only approximately 21 million of these are kept for training due to various consistency checks applied to the code outputs. More information about the QuaLiKiz code can be found at www.qualikiz.com. The data is saved under 3 keys in HDF5 format: "/input", "/output", and "/label". The inputs to the QuaLiKiz evaluations are provided under "/input", representing the local plasma parameters extracted from experimental measurements from the JET plasma device in Culham, UK, along with variations of select parameters according to propagated experimental uncertainties. Selected relevant outputs of the QuaLiKiz evaluations are provided under "/output", namely the local turbulent transport coefficients after applying a semi-empirical turbulent fluctuation saturation rule. Some useful metadata is provided under "/label", giving some degree of provenance tracking back to the JET experimental database, as well as describing the applied parameter variations and explaining why certain output rows were removed from the output structure.本数据集用于训练QuaLiKiz神经网络（QuaLiKiz-neural-network，简称QLKNN）模型QLKNN-jetexp-15D，相关细节刊载于下述已发表学术论文：https://doi.org/10.1063/5.0038290。该数据集通过对QuaLiKiz-v2.6.2执行约3300万次独立评估生成，每次评估均采用标准18维波数向量。由于需对代码输出开展多项一致性校验，最终仅保留约2100万条数据用于模型训练。有关QuaLiKiz代码的更多信息可访问其官方网站www.qualikiz.com。数据以HDF5格式存储，包含三个数据键："/input"、"/output"与"/label"。其中，"/input"项下存储QuaLiKiz评估的输入数据，即从英国卡勒姆JET等离子体装置的实验测量结果中提取的局域等离子体参数，同时包含依据传播的实验不确定度得到的选定参数变体。"/output"项下存储QuaLiKiz评估的选定输出结果，即应用半经验湍流涨落饱和规则后的局域湍流输运系数。"/label"项下附带若干实用元数据，可实现一定程度的溯源追踪，回溯至JET实验数据库，同时说明所施加的参数变更规则，并解释为何部分输出行从输出结构中被移除。