Prediction of the dielectric dissipation factor tan δ of polymers with an ANN model based on the DFT calculation

A quantitative structure-property relationship (QSPR) model was constructed to predict the dielectric dissipation factor (or power factor or electrical loss tangent) tan δ of polymers by means of artificial neural network (ANN). The frequency of measurement (ν) and five quantum chemical descriptors (qR+, q−(R/M), EMLUMO, EM/RLUMO, and SR) calculated at the DFT/B3LYP/6-31G(d) level were used as vectors to develop the model. The typical back-propagation (BP) neural network was employed for fitting the possible non-linear relationship existed between the six descriptors and tan δ. The optimal condition of the neural network was obtained by adjusting various parameters by trial-and-error. Simulated with the final optimum BP neural network [6-2-1], the results show that the predicted tan δ values are in good agreement with the experimental ones, with the root mean square error (rms) being 0.01067 (R = 0.939) for the training set and 0.01463 (R = 0.902) for the test set. Comparing with existing models, the model proposed is independent of the refractive index n and the dielectric constant ε. Thus the present model is more useful in predicting the tan δ values for polymers.