An ANN model was designed to predict unstable states in MILD combustion systems out of six kinds of input factors. The effectiveness of the established ANN model was validated, demonstrating accurate predictions for the imbalanced classification problem in systems described by both GRI3.0 and POLIMI2003 mechanisms. The predictions in high-dimensional parameter spaces revealed that unstable states are more likely to occur under stoichiometric conditions or in the presence of a reactive bath gas, such as CO or HO. Additionally, these states could manifest in narrow parameter spaces, such as within a very confined mid-temperature range in a fuel-rich system with a low dilution level. Interestingly, the analysis of dominant reactions and feedback loops unveiled similarities in thermodynamic feedback mechanisms across a spectrum of parameter combinations. Meanwhile, feedback loops construct shortcut pathways on the level of oxidation extent and can facilitate the switching between high and low temperature chemistry.

中文翻译：

---

人工神经网络辅助不稳定燃烧状态预测和主导化学动力学分析


ANN 模型旨在根据六种输入因素预测 MILD 燃烧系统的不稳定状态。验证了所建立的ANN模型的有效性，证明了对GRI3.0和POLIMI2003机制描述的系统中不平衡分类问题的准确预测。高维参数空间中的预测表明，在化学计量条件下或存在反应浴气体（例如 CO 或 H2O）的情况下，更可能发生不稳定状态。此外，这些状态可能会在狭窄的参数空间中显现出来，例如在低稀释水平的富燃料系统中非常有限的中间温度范围内。有趣的是，对主要反应和反馈回路的分析揭示了一系列参数组合的热力学反馈机制的相似性。同时，反馈回路在氧化程度水平上构建了捷径，可以促进高温和低温化学之间的切换。