Machine learning-driven multi-objective optimisation of ammonia co-firing with highly reactive fuels

• A modelling framework was developed to investigate ammonia co-firing with highly reactive renewable fuels. • Machine learning models were developed to efficiently determine optimal operating conditions. • Multi-objective optimisation revealed trade-offs between combustion performance and emissions. • Equivalence ratio and fuel composition played key roles in determining combustion characteristics. • Conditions for achieving burning velocity comparable to conventional fuels with lower emissions were identified. Co-firing ammonia with highly reactive fuels, especially renewable fuels such as hydrogen, dimethyl ether, and methanol, can effectively enhance ammonia combustion efficiency and stability, while aligning with the global transition toward sustainable energy solutions. However, this strategy leads to emissions of pollutants (NOx, N 2 O, and CO). In this study, we developed a framework based on machine learning and multi-objective optimisation methods to search for optimal ammonia co-firing operating conditions, aiming to enhance key combustion outputs while minimising emissions. First, we performed one-dimensional premixed flame simulations to generate data across a wide range of combustion conditions. Using this dataset, we evaluated the performance of nine machine learning models and identified the multi-layer perceptron (MLP) model as the most effective one. The MLP model exhibited high efficiency and accuracy on the test data, but also a satisfactory ability to generalise, allowing the trained models to yield accurate predictions under out-of-sample operating conditions. The SHapley Additive exPlanations analysis was applied to understand the influence of different input variables on the MLP model outputs, thereby enhancing its interpretability and reliability. Finally, we integrated the MLP model with the Non-dominated Sorting Genetic Algorithm-III (NSGA-III) algorithm to generate Pareto front solutions for different operating conditions. The Pareto front solutions provide a set of trade-offs between several conflicting objectives in the combustion process. Using multi-criteria decision-making methods, we finally identified the optimal conditions for the co-firing systems, improving the combustion efficiency and dwindling emissions. Furthermore, this versatile framework can be extended to more complex combustion systems, offering an efficient solution for optimising combustion processes.