Air quality data prediction method and system

Abstract

The invention discloses an air quality data prediction method and system, and the method comprises the steps: carrying out the collection and preprocessing of multi-source data, screening key feature variables through employing a CatBoost algorithm, and carrying out the signal decomposition through combining with an NA-RPSEMD method, and obtaining an IMF component; main variable weighting is carried out on the decomposed IMF components, the contribution degree of each component is calculated, and optimized feature components are obtained; optimizing a Kepler algorithm KOA, improving a population initialization process by adopting a boundary mapping optimized good point set strategy and a reverse learning strategy, and introducing a tangent function to adjust an adaptive weight to obtain an improved Kepler optimized MKOA algorithm; an STGAMform air quality prediction model is established, an MKOA algorithm is utilized to optimize hyper-parameters of the STGAMform and weights corresponding to main variable weighting, air quality data are predicted, and an air quality prediction result is output; according to the method, key features can be automatically identified, data noise can be effectively suppressed, and the accuracy and robustness of feature extraction are remarkably improved.

Images