Abstract: For the purpose of improving the accuracy of predicting earthquake elements, basic core functions of various types were built on the same network model and multi-core neural network integration models were built for different types of radial basis functions to improve the accuracy of the network. The optimization was conducted from determining the amount of optimal radical basis neurons, increasing the target error of training appropriately and other aspects to reduce the minimum training error and increase the accuracy of prediction. Multiple regression analysis method was adopted to fit the samples so as to obtain the multiple regression coefficients of sub-predictions; the same method was adopted to conduct multiple regression integration on the sub-prediction models and perform anti-normalization treatment on the data, and eventually obtained the regression-integrated multi-core RBF (radical basis function) prediction model. The results indicated that the model of multiple regression prediction integration established by this multi-core neural network integration method for prediction of earthquake elements is capable of achieving the optimal fitting of actual value and predictive value as well as earthquake element predictive values of higher accuracy.