Abstract: PM2.5 concentration is an important indicator for assessing ambient air quality, so accurately predicting the change trend of PM2.5 concentration will help to formulate more effective environmental protection measures. In the paper, the Sparrow Search Algorithm (SSA) is employed to determine the optimal values of regularization function and kernel function parameters of Kernel Extreme Learning Machine (KELM), thereby constructing the SSA-KELM model, and then the improved SSA-KELM model is utilized to predict PM2.5 concentration. Based on the air pollutants and meteorological data of Hefei City, Anhui Province, China, this study firstly uses Pearson coefficient to evaluate the correlation between other factors and PM2.5 concentration, then uses stepwise regression method to screen out the factors with high correlation with PM2.5 concentration to input into the SSA-KELM model, and finially achieves the prediction of the daily average concentration of PM2.5 using the SSA-KELM model. According to the prediction results, the SSA-KELM prediction model exhibits superior performance in accuracy and generalization ability, with a mean square reduced to 0.909 and a fitting degree of 0.998, indicating that the constructed SSA-KELM prediction model has good prediction ability for the change trend of PM2.5 daily average concentration.

Key words: sparrow search algorithm, kernel extreme learning machine, PM2.5, prediction, stepwise regression