A model to predict property of additives modified carbon material high temperature binder with RBF neural networks

On the basis of experimental data about carbon material binder modification additives, the bond strength prediction model of the carbon material with RBF NN (radial basis function neural network) is studied. An improved hybrid algorithm of nearest neighbor clustering algorithm (NNCA) and mode 2 decreasing gradient descent (M2DGD) is proposed to solve the low accuracy problem of NNCA. Then the prediction accuracy and the training process between NNCA RBF NN, NNCA-M2DGD RBF NN and BP (back-propagation) NN are compared. The results showed that the average relative errors of these three models are 0.0127, 0.0113 and 0.0622 respectively. The RBF neural network prediction model is the best. Finally the optimal formula is estimated. The RBF NN using this improved algorithm is very suitable for learning functions from experimental data and has efficient ability of prediction. Therefore, RBF NN is expected to use in multivariable, nonlinear system such as the carbon material binder modification additives. RBF NN is a kind of prospect theoretical design methods for carbon material.