Abstract. Utilization of PM 10 concentrations and meteorological data of Ningdong Base, the PM 10 concentration is predicted based on LS-SVR, BP-ANN and traditional MLR models. The results show that the LS-SVR model can better describe the nonlinear dependence between PM 10 concentration and meteorological factors, and predict the PM 10 concentration more accurately.

Abstract-We select the Ningdong Energy and ChemistryIndustry Base (referred to as Ningdong Base) as research object and take the ecological environment as evaluating target. According to the subject and object factors, the SVM model of ecological environment vulnerability assessment is constructed by using principal component analysis method. We embed GA in SVM to optimize its structure and parameters. SVM model is simple, universal, and accurate, and it can be applied in evaluation of eco-environmental vulnerability in practice. Also, we compare our evaluation results with the the BP-ANN model's results. Then, we draw a conclusion that the ecological environment vulnerability of the Ningdong Base is the second level, which belongs to moderate vulnerability. Finally, the causes of vulnerability formation are analyzed.