Abstract: The Radial Basis Function (RBF) neural network was used to predict the contents of soil heavy metals of Cr, As, Hg, Ni, Pb, Zn in study area. We selected 35 groups of measured data as training data, with other 6 groups as validation data. The model was constituted continuous data of the 5 groups followed 10 groups of data based on continuous sampling. The number of neurons in the input layer is 10, and the output layer is 5. The transfer function of the hidden layer is the radial basis function radbas, and the transfer function of the output layer is linear function Purelin. The results showed that:RBF neural network prediction model has higher precision., Through multivariate statistical analysis on measured samples and prediction samples, the average values of As, Ni, Zn in the study area were found to exceed the background values of the soil environment in Shanghai City, and the As elements reached a high degree of variation, and Pb, Zn and Ni reached moderate variation. Through factor analysis, the first 2 factors basically contain the main information of all the elements. Among the first factors, the maximum load is the element Ni (0.946), and the second factor is the element As (0.930). Through the evaluation of potential ecological risk index, the study area is found to in a slight ecological risk level. The RBF neural network model can be used to reduce the cost of sampling and analysis, and to evaluate the ecological risk of heavy metals in the regional soil.

Key words: RBF neutral network, heavy metals, prediction, ecological risk