Parameter optimization of BP-neural network based on the forecast of cast blasting

A method of optimizing parameters of BP cast blasting model was described.The accuracy of predicting result and BP model convergence rate with different numbers of hidden layers,different training functions and transfer functions,different learning rates were analyzed particularly.The forecast result of parameter optimization model was contrasted with those of RBF(Radial-Basis Function),SVM(Support Vector Machine) models.The experiment was done by using a lot of surveying data of some open pit cast blasting.The result indicates that the optimizing BP topology structure is 10-6-3(10 input variables,6 hidden layers,3 output variables);LM(Levenberg-Marquart) algorithm is the best training function;tansig and purelin functions combination is the best transfer function;the best learning rate is 0.77;test sample simulation standard deviation of cast blasting ratio,max cast distance,coefficient of volumetric expansion based on parameter optimization BP model is 9.567 8,0.036 3,0.041 4 respectively and is the smallest contrasting with RBF and SVM;the prediction result of parameter optimization BP model is the best one.