Combining genetic algorithm and deep learning to optimize a chemical kinetic mechanism of ammonia under high pressure
Combining genetic algorithm and deep learning to optimize a chemical kinetic mechanism of ammonia under high pressure
复制标题
DOI：
10.1016/j.fuel.2023.130508
复制
发表时间：
2024-03
期刊：
影响因子：
7.4
通讯作者：
Long Liu;Fusheng Tan;Zan Wu;Yang Wang
中科院分区：
文献类型：
--
作者：
Long Liu;Fusheng Tan;Zan Wu;Yang Wang
氨是一种潜在的零碳燃料，但高压实验的困难限制了其在电力工程中的应用发展。本研究将深度学习与遗传算法相结合，对高压下氨的化学反应动力学机理进行了优化。结果表明，深度学习模型能够在小于0.2% |Elog|（对数绝对误差）的范围内回归点火延迟时间的实验数据。同时，训练后的模型可用于高压下点火延迟时间的趋势预测，丰富了实验无法获得的氨燃料燃烧数据。采用强化精英遗传算法对大种群规模的PLOG反应进行优化，最终将氨化学反应动力学机理的预测精度提高了约5%。论证了人工智能在燃烧科学中的应用价值。
Ammonia is a potential zero-carbon fuel, but the difficulties of experimenting at high pressure limit its development of application in power engineering. In this work, deep learning and genetic algorithm are combined to optimize the chemical reaction kinetics mechanism of ammonia under high pressure. The result shows that deep learning model is able to regress experimental data of ignition delay time in the range of less than 0.2% |Elog| (logarithmic absolute error). At the same time, the trained model can be applied to the trend prediction of ignition delay time at high pressure, enriching combustion data of ammonia fuel that is not available from experiments. Optimization of PLOG reactions with a large population size was performed by strengthen elitist genetic algorithm, and the prediction accuracy of ammonia chemical reaction kinetics mechanism was improved by about 5% finally. The application value of artificial intelligence in combustion science is demonstrated.