Prediction of polycarbonate degradation in natural atmospheric environment of China based on BP-ANN model with screened environmental factors

The degradation of polycarbonate (PC) varies with different service environments. To predict the degradation of PC in atmospheric environments, a back propagation artificial neural networks (BP-ANNs) model was constructed based on datasets from long-term exposure tests in 13 representative cities of China. Based on the analysis by Pearson correlation method and factor analysis, as well as the ranking of environment parameters that influence degradation performance by grey correlation method, 4 key environment parameters were identified. To obtain the optimized model, BP-ANNs with different input environmental parameters, middle layers and training precision were compared. The high prediction accuracy and generalization ability of the well trained BP-ANN were verified using datasets from atmospheric weathering conducted in three new locations. Furthermore, a high-resolution predictive map for PC degradation was drawn based on the yellow indices which were predicted by inputting the 4 key environment parameters of 804 cites in China. Results showed PC degrades most seriously in the tropical monsoon climate area and plateau climate area, but slightly in the temperate monsoon climate area in the northeastern, and subtropical monsoon climate area in the southwestern basin of China. The model developed in this study would benefit to the rapid design, selection and evaluation of PC-based components in atmospheric service environments.