A Novel Coupled Optimization Prediction Model for Air Quality

Abstract

PM2.5 is a significant pollutant that negatively affects atmospheric environmental sustainability, and accurate prediction of its concentration is crucial. Most existing prediction models face challenges such as inadequate data feature capture, dismissal of influential factors, and subjective model parameter tuning. To address these issues, this paper introduces a novel coupled air quality optimization prediction model based on Variational Mode Decomposition (VMD), the Informer time series algorithm, Extreme Gradient Boosting (XGBoost), and the Dung Beetle Optimization Algorithm (DBO). The coupling approach screens influential features using the Spearman coefficient method, optimizes VMD with DBO, decomposes time series data, and classifies various feature data according to approximate entropy. The Informer algorithm and DBO-optimized XGBoost process different feature data separately, then superimpose and reconstruct the predicted values to obtain results. Using air quality prediction in Nanjing as an example, the new model achieves superior performance (R-squared=0.961, RMSE=1.988, MAE=1.624). Compared to the WANNs model with the highest accuracy in recent relevant studies, our model demonstrates a 2.96% increase in R-squared, a 21.89% decrease in RMSE, and a 20.05% decrease in MAE. This comparison illustrates that the proposed DBO-VMD-Informer-XGBoost prediction model effectively addresses the limitations of existing air quality prediction models and offers increased prediction accuracy. By employing the advanced DBO algorithm for prediction and innovatively combining VMD, Informer, and XGBoost, this model presents high potential in air quality prediction and is anticipated to have broader applications.