Solubility is one of the most important physicochemical properties of polymer compounds, which determines the compatibility of blending system. To enhance the performance of artificial neural networks (ANN) and improve the efficiency and correlation of prediction of gas solubility in polymers, in this work, a novel ANN model based on chaos theory, self-adaptive particle swarm optimization (PSO) algorithm and back propagation (BP) algorithm is proposed, hereafter called CSAPSO-BP ANN. In the CSAPSO-BP ANN, the conventional PSO algorithm is modified by using chaos theory and self-adaptive inertia weight factor to overcome its premature convergence problem. Then the CSAPSO-BP ANN trained by hybrid algorithm which combined the modified PSO and BP algorithm has been employed to investigate carbon dioxide (CO₂) solubility in polystyrene (PS), polypropylene (PP) and nitrogen (N₂) solubility in PS, respectively. The CSAPSO-BP ANN model which consisted of three layers with one hidden layer, two input nodes including temperature and pressure, 8 hidden nodes which obtained by heuristics and one output node that is the solubility of gases in polymers was designed. The model combined the abilities of chaos theory, PSO algorithm and BP algorithm, accelerated the training speed of ANN and improved the prediction accuracy. Results obtained in this work indicate that the CSAPSO-BP ANN is an effective method for prediction of gas solubility in polymers in a wide range of pressure and temperature. The comparison between different neural networks was carried out in detail to reveal the proposed CSAPSO-BP ANN outperforms the traditional BP NN and PSO-BP NN. The values of average absolute deviation (AAD), standard deviation (SD) and squared correlation coefficient (R²) are 0.0058, 0.0198 and 0.9914, respectively. The statistical data demonstrate that the CSAPSO-BP ANN model is a faster, more reliable and accurate method, and has an excellent prediction capability with high-accuracy and has a good correlation between prediction values and experimental data.

Key words: solubility prediction, polymer, neural network, particle swarm optimization, chaos theory