关键词: BP神经网络, 遗传算法, 自适应交叉变异, 最优保存, 物性预测, 熔点

Abstract: In order to avoid local minimum points during training BP network,adaptive crossover mutation algorithm and elitist model algorithm were used,which could optimize the weight and bias of BP network so as to build a hybrid genetic algorithm neural network. The algorithm consisted of the respective advantages of adaptive crossover mutation algorithm and elitist model algorithm. The best solution of problems was obtained by using as few generations as possible.In this way, local minimum point was prevented and the better average fitness and individual of best fitness were obtained in the algorithm. The artificial neural network with the above optimization strategy could greatly improve stability and speed of convergence and guarantee convergence to a global minimum point.Application of artificial neural network to predict physical properties of materials is a promising field to be developed. In this paper the proposed artificial neural network based on hybrid genetic algorithm was used to predict melting points of organic compounds. The molecular mass, critical density, normal boiling point and dipole moment of organic compounds were used in prediction.The predicted values based on artificial neural network with hybrid genetic algorithm were better than the values calculated with the Joback equation and Xu equation.

Key words: BP神经网络, 遗传算法, 自适应交叉变异, 最优保存, 物性预测, 熔点