Prediction of Excess Air Ratio Through Deep Neural Network–Based Multidimensional Analysis of OH∗ Radical Intensity and Fuel Pressure in Flame

Abstract

This study proposes a deep neural network (DNN)–based regression model for predicting the excess air ratio, which is a critical indicator for optimizing combustion efficiency and minimizing harmful emissions in industrial combustion systems. The chemiluminescence signals of the OH^∗ radicals and fuel pressure were used as the input features for the prediction model. To evaluate the effect of the multidimensional input, Case 1, with only the OH^∗ radical signal as a single input, was compared with Case 2, with the OH^∗ radical signal and fuel pressure as the inputs. The results showed that the Case 2 model reduced the mean absolute error (MAE), mean relative error (MRE), and root mean squared error (RMSE) by approximately 40.71%, 41.85%, and 19.69%, respectively, compared to Case 1, and the average relative prediction error rate was also 2.25% lower. These results demonstrate the potential for improving the accuracy and generalization ability of the model by incorporating multidimensional input features. Therefore, DNN models using multidimensional inputs can contribute to the design and implementation of combustion control systems to optimize the combustion efficiency and reduce harmful emissions in industrial combustion systems by predicting the excess air ratio.