Metaheuristics Optimized Machine Learning Modelling of Environmental Exergo-Emissions for an Aero-Engine [2022]

Abstract

The purpose of this study is to develop a metaheuristic design for primary parameters and architectures of two models of artificial neural network (ANN) in predicting a cargo aircraft's exergo-emissions (exergy destruction ratio, r(ex,dest) , and waste exergy ratio, r(wex) ) at different flight stages. Hybrid genetic algorithm (GA)-ANN models have been accomplished utilizing real databases of r(ex,dest) and r(wex) at various powers. Implementing a metaheuristics-based optimization on multilayer perceptron (MLP)-ANNs has produced the most favourable initial weights, step-size, biases, and training algorithm's back-propagation (BP) momentum rate in addition to optimum number of neurons in the hidden layer(s). In accordance with an error assessment, a close fit linking real data and r(wex) (linear correlation ratio, R, value of 0.999851) as well as r(ex,dest) (R value of 0.999985) predicted values is found. In the r(ex, dest) estimation model, the accuracy among single-hidden-layer networks has been confirmed to be higher; whereas, highly accurate testing outcomes have been obtained in two-hidden-layer networks as far as modeling of r(wex) is concerned. ANN models' optimization by GAs has increased the accuracy of the resulting models (R value of 0.999987 and 0.999869 for r(ex,dest) and r(wex) , in that order ascertaining a drop-off in the testing stage errors).