Design of a combined system based on multi-objective optimization for point and interval forecasting of air pollution

Serious air pollution poses destructive effects on environmental safety and human health, which is a public threat worldwide. Therefore, it is important to develop a reliable air pollution forecasting system to monitor the air quality in advance. Most of the existing researches lack data feature mining and uncertainty analyses of the predictions, leading to insufficient results. This study proposes a novel forecasting system that comprises a hybrid data preprocessing-analysis module, a combined deterministic prediction module, and a probabilistic forecasting module to overcome the above-mentioned drawbacks. Specifically, the air pollution data are decomposed and reshaped to eliminate negative disturbances to achieve high-quality data input for forecasting. Then, four individual models and a multi-objective weight determination strategy are combined to achieve the point forecasting results. After obtaining fitting errors, their distributions are analyzed to achieve forecasting intervals under different confidence levels. Finally, twelve datasets from three cities in China were employed for experiments and the obtained results have shown that the proposed model achieves more accurate and stable predictions than other benchmark models, providing reliable and comprehensive information for monitoring air quality. © 2021 Elsevier Ltd