Global Warming Potentials (GWPs) of refrigerants and related compounds are calculated by complex simulation and periodically updated by World Meteorological Organisation (WMO). A new attempt has been made to predict GWPs of such compounds based on their molecular structure using Artificial Neural Network (ANN) and Group Contribution Method (GCM). The main focus is on refrigerants, including CFCs, HCFCs, HFCs, HFOs, HCs, chlorocarbons, fluorocarbons, bromocarbons, iodocarbons, oxygenated and unsaturated halocarbons. In order to give some statistical advantage to the ANN model, many other related families of compounds, totally about 495 compounds, have also been considered. Only GWPs for the time horizon of 100 years are used in this analysis. Considering the complexities of the problem and the confidence limits on GWP data published by WMO (2018), the proposed ANN with GCM model is able to predict the GWP within reasonable accuracy. A comparative analysis with another complex prediction method establishes the superiority of this direct GWP predictions from the molecular structure of a compound.

制冷剂和相关化合物的全球变暖潜能值 (GWP) 是通过复杂的模拟计算得出的，并由世界气象组织 (WMO) 定期更新。使用人工神经网络 (ANN) 和群贡献法 (GCM) 根据分子结构预测此类化合物的 GWP 进行了新的尝试。主要关注制冷剂，包括 CFCs、HCFCs、HFCs、HFOs、HCs、氯烃、碳氟化合物、溴烃、碘烃、含氧和不饱和卤代烃。为了给 ANN 模型提供一些统计优势，还考虑了许多其他相关的化合物家族，总共大约 495 种化合物。本分析仅使用 100 年时间范围内的 GWP。考虑到问题的复杂性以及 WMO（2018 年）发布的 GWP 数据的置信限制，提出的具有 GCM 模型的 ANN 能够在合理的准确度内预测 GWP。与另一种复杂预测方法的比较分析确定了这种从化合物分子结构直接预测 GWP 的优越性。