MetaForecaster: A PSO-Driven Neural Model for Sustainable Industrial Air Quality Management

Abstract

Industrial carbon monoxide (CO) emissions significantly affect public health and environmental quality, necessitating advanced forecasting models for effective air quality management. Traditional neural network (NN)-based forecasting methods frequently exhibit limitations, including inadequate hyperparameter tuning and limited responsiveness to temporal variability in industrial emissions data. To address these challenges, this study proposes an optimized neural forecasting framework integrating Particle Swarm Optimization (PSO) with neural networks. The PSO algorithm strategically optimizes network weights and biases, utilizing the mean squared error (MSE) as the fitness metric to ensure prediction accuracy. The framework is validated using a segmented real-time dataset that distinguishes daytime and nighttime CO emissions, improving the adaptability and precision of the model. Comparative analyzes with established hybrid forecasting approaches, such as Genetic Algorithm-NN, Simulated Annealing-NN, and Differential Evolution-NN, demonstrate the superior performance of the proposed PSO-NN model, achieving notably low prediction errors (MSE: $1.1941 \times 10^{-7}$ ), MAPE: 0.0016 and a high coefficient of determination ( $R^{2}$ : 0.99999). Furthermore, Theil’s U statistic confirms the robustness and predictive reliability of the model. Consequently, the proposed PSO-NN framework emerges as an effective real-time decision support system, facilitating sustainable air quality governance and promoting environmentally responsible industrial production practices.