The formation of haze depends on the complex evolution behavior of SO 2 , NO 2 , and CO. We explore the influence of surrounding urban oxides on haze formation in Beijing. From the perspective of time evolution, multifractal detrended cross-correlation analysis (MF-DCCA) method quantitatively shows that all oxides reveal persistent cross-correlations (h xy : 0.640 sim; 0.995) in the short term and long term. The correlation characteristics of the same oxide in different regions are compared. In the short term, Zhangjiakou (SO 2 /NO 2) and Tianjin (CO) have the strongest multifractal features, while Zhangjiakou (SO 2 /NO 2 /CO) is more prominent in the long term. Their sources are mainly Fat-tailed. Ignoring the characteristics of timing, the fitting degree (R 2) of LightGBM model to PM2.5 concentration regression is 0.862 after the addition of neighboring oxides, an increase of 2.0%. CatBoost's R 2 improves by 6.4%. The feature importance score indicates that Langfang's SO 2 and Chengde's CO contribute the most to the formation of PM2.5. This study has certain reference value for the formulation of oxide emission strategy in the surrounding area of Beijing.