BP网络泛化能力改进方法在染色配色中的应用研究

针对BP神经网络在织物染色配色中的局限性,提出了对BP神经网络应用到织物染色配色中的泛化能力改进方法。实验及仿真结果表明,基于泛化能力改进的BP神经网络在平均均方差上优于标准的BP神经网络,并且误差函数改进的BP神经网络要优于网络训练提前结束法改进的BP网络,与一般的BP神经网络模型相比,计算精度有了较大的提高。

Research on the Method for Improving the Generalization Ability of BP Neural Network Used in Textile Dyeing Color Matching

In order to solve the limitations problem of BP neural network used in color matching for textile dyeing,a method to improve the generalization of BP neural network used in color matching for textile dyeing is proposed in this paper. Experiments and simulation results show that the average mean square deviation based on improved generalization of BP is better than the one based on standard BP. And the BP based on improved error function is superior to the one based on the stopping training ahead of schedule method. Compared with the general BP algorithm,the calculating accuracy of the method proposed in paper increases considerably.