【摘要】 本文采用二维直接数值模拟方法，研究了H₂/空气在不同湍流强度下的火核形成和发展过程，基于此建立了直接数值模拟数据库。数值结果表明，湍流强度对燃烧过程呈现双重效应：湍流通过增强局部混合效率和火焰面褶皱来加速燃烧进程，但湍流强度过高会导致火焰前沿断裂、高温区域分布不均匀，破坏燃烧稳定性。之后搭建CNN-LSTM神经网络模型对不同强度下H₂/空气燃烧的温度场、OH质量分数和放热率的空间分布开展了预测，与DNS结果进行对比并采用性能评价指标进行验证。结果显示，模型预测结果与DNS结果吻合良好，决定系数R²大于0.98，峰值信噪比大于40 dB，说明所建立的CNN-LSTM模型可以提取时间和空间特征，实现了对H₂/空气火核燃烧结构演变的精准预测。更多还原

【Abstract】 Two dimensional direct numerical simulations were performed to study the formation and development process of hydrogen/air flame kernel under different turbulent velocities, and then establishes a direct numerical simulation database correspondingly. Enhanced turbulent motion improves local mixing efficiency and induces flame surface wrinkling, accelerating combustion progression, while excessive turbulence leads to flame front fragmentation and spatially heterogeneous high-temperature zones, thereby compromising combustion stability. Subsequently, a CNN-LSTM neural network model was built to predict the temperature field, OH mass fraction, and heat release rate of hydrogen/air combustion under different turbulent velocities. Quantitative validation was performed against high-fidelity DNS datasets using two specific metrics: the coefficient of determination(R~2) and peak signal-to-noise ratio(PSNR). The model predictions show strong agreement with DNS results, demonstrating R~2 values exceeding 0.98 and PSNR above 40 dB. These results confirm that the CNN-LSTM model effectively captures spatiotemporal features to enable accurate prediction of hydrogen/air flame kernel structural evolution.更多还原