Prediction of energy consumption and airflow of a ventilation system: A SAGA-optimised back-propagation neural network-based approach

Abstract

With the rapid advancement of computational technologies and machine learning algorithms, predicting airflow in underground mine ventilation systems has become increasingly feasible. However, some algorithms present challenges due to low convergence rates and susceptibility to getting trapped in local minima, such as the back-propagation neural network (BPNN). This paper introduces a novel approach that combines the strengths of a global search genetic algorithm (GA) and a local search simulated annealing (SA) algorithm, referred to as the SAGA method, to address these limitations. The SAGA method focuses on optimising the initial weights and thresholds of the BPNN, effectively mitigating the issue of rapid convergence into local minima. A key innovation is incorporating an adaptive learning rate into the BPNN algorithm, resulting in an improved SAGA-BP prediction model. This model forecasts airflow within underground mine ventilation systems in a laboratory-based prototype. Experimental results testify to the efficacy of the SAGA-BP model. Compared to the conventional SAGA-BP model, the proposed approach consistently demonstrates higher accuracy with (R2 of 0.941 and test MSE of 0.3513) and faster underground mine ventilation airflow prediction. Thus, this approach can revolutionise mine ventilation and monitoring technologies by lowering energy consumption and operational costs, increasing mine productivity, improving system performance and reliability, and, most importantly, improving health and safety. © 2025 Elsevier Ltd