Multi-objective optimisation of the emission parameters and efficiency of pellet stove at different supply airflow positions based on machine learning approach

Abstract

The main aim of this study is to optimise the effect of burner pot design by changing supply airflow position in a pellet stove. For this purpose, the calibration model based on Gaussian Process Regression (GPR) with optimised hyper-parameters was proposed to determine the relationship between the input and output parameters of the pellet stove. The designed model was validated by the experimental data and was measured by several statistical tools. The modelling findings showed that the developed model was an efficient technique and has the potential to predict pellet stove output characteristics with high accuracy. Subsequently, a multi-objective optimisation (MOO) by genetic algorithm using this developed GPR model was implemented to design the burner pot. The optimisation process was repeated for different excess air ratios with manufactured different burner pots which have four different supply airflow distances and the performance of the burner pots with optimised geometries was compared to the average of experimental data in three total thermal power input cases. The results showed that the significant improvements on emission and efficiency performance of the pellet stove was achieved with the optimised inputs in both each thermal power inputs and global cases.