Machine learning based heat release rate indicator of premixed methane/air flame under wide range of equivalence ratio

Abstract

Heat release rate (HRR) is an essential parameter of describing and monitoring combustion phenomenon, which has intricate spatial and temporal distributions. Proper evaluation of the performance of combustion systems require accurate determination of local HRR, while it cannot be directly measured in practical combustion field. At present, it's still a difficult issue to find a generally-accurate HRR indicator for wide range of equivalence ratios. In this work, machine learning methods are adopted to find and construct HRR indicator of premixed methane/air flames at lean-to-stoichiometric condition. First, Typical conventional HRR indicators are evaluated. Then, three types of machine learning methods, artificial neural network algorithm, support vector regression, multiple linear regression, are used to construct new HRR models and the accuracies are evaluated. Multiple linear regression algorithm is ultimately recommended for constructing HRR models, due to its high prediction accuracy, the lowest model complexity and simplicity of parameter adjustment. Finally, a third-order multiple linear regression model based on radical CH3 and O is proposed and recommended as HRR indicator, which has high accuracy under different temperature and lean-to-stoichiometric condition.