Abstract
Dual-fuel NH3/CH4 gas turbines are often operated under high-temperature and rich-fuel conditions (such as staging combustion) in order to reduce both carbon and NOx emissions simultaneously. The environment alters the oxygen concentration and the way in which free radicals are generated. These factors make it difficult to predict ignition delay quickly and accurately. The present study proposes a novel CNN-APC-DML model. The novel model maintains an R2 value of 0.88-0.90 on the test set under operating conditions for a full fuel component at temperatures of 1200-1800 K, pressures of 1-5 atm and equivalent ratios of 1.0-1.6 once the extrapolation capability is verified. The maximum MAE and RMSE are only 0.053 and 0.081 respectively. Its computational time for forecasting tasks involving the same volume of data is 99.42 per cent lower than numerical simulation software. Multivariate analyses including Pearson, SHAP and DML reveal the temperature is the primary factor influencing the prediction results. The equivalent ratio has a greater net effect than pressure. The new model provides a reliable tool for the rapid prediction of NH3/CH4 ignition delay under high-temperature and rich-fuel conditions. It also offers new insights into the physical-causal co-modelling of complex combustion processes.

This publication has 20 references indexed in Scilit: