While there have been various attempts in establishing a model that can predict the flammability characteristics of polymers based on their molecular structure, artificial intelligence (AI) presents a promising alternative in tackling this pressing issue. Therefore, a novel approach of adopting AI methods, extreme learning machines (ELMs) and group method of data handling (GMDH) in estimating heat release capacity, total heat release and char yield from thermophysical properties of polymers were addressed. GMDH showed a clear indication of overfitting whereby the models generated excellent training results but could not provide similar performance during testing. The superior generalisation performance of ELM during testing makes it the standout method. ELM produced HRC predictions having R and RRMSE of 0.86 and 0.405 for training, 0.94 and 0.356 for testing. For THR estimates from ELM, the R and RRMSE scores were 0.9 and 0.195 for training, 0.93 and 0.131 for testing. While char yield ELM model generated 0.88 and 0.795 for training, 0.93 and 0.383 for testing. The potential of ELM was demonstrated as it estimated the flammability parameters of 105 polymers having little or no empirical test results.

中文翻译：

---

开发用于预测燃烧和可燃性特性的人工智能模型

尽管在建立可以根据聚合物分子结构预测聚合物可燃性特征的模型方面进行了各种尝试，但人工智能 (AI) 为解决这一紧迫问题提供了一种有希望的替代方案。因此，提出了一种采用 AI 方法、极限学习机 (ELM) 和数据处理组方法 (GMDH) 来估计聚合物热物理特性的热释放能力、总热释放和炭产量的新方法。GMDH 显示出明显的过度拟合迹象，模型产生了出色的训练结果，但在测试期间无法提供类似的性能。ELM 在测试期间卓越的泛化性能使其成为出色的方法。ELM 产生的 HRC 预测具有 0.86 和 0.405 的 R 和 RRMSE 用于训练，0.94 和 0。356 进行测试。对于来自 ELM 的 THR 估计，训练的 R 和 RRMSE 得分分别为 0.9 和 0.195，测试的 R 和 RRMSE 得分分别为 0.93 和 0.131。而 char yield ELM 模型生成 0.88 和 0.795 用于训练，0.93 和 0.383 用于测试。ELM 的潜力得到了证明，因为它估计了 105 种聚合物的可燃性参数，几乎没有或没有经验测试结果。