论文部分内容阅读
主要研究动量BP算法、学习率可变的动量算法等2种方法分别对苋菜红、日落黄、柠檬黄、胭脂红等4种合成色素混合体系的同时预测,并对计算效果进行比较。学习率可变的动量算法的参数及结果如下:当隐含层节点数为S=6,学习率0.21,动量因子η=0.6,RPEt=0.27105,RPEs分别为0.35216,0.31452,0.28921,0.28718,训练步数142次就能够达到预定精度。结论:学习率可变的动量算法比动量BP算法速度快,准确率高。
Mainly studied the two methods of momentum BP algorithm and variable momentum rate method to predict the four kinds of synthetic pigment hybrid systems, such as amaranth, sunset yellow, lemon yellow and carmine respectively, and compare the calculated results. The parameters and results of the momentum algorithm with variable learning rate are as follows: when the number of hidden layer nodes is S = 6, the learning rate is 0.21, the momentum factor η = 0.6, RPEt = 0.27105, RPEs are 0.35216,0.31452,0.28921,0.28718, respectively 142 steps to reach the predetermined accuracy. Conclusion: The momentum algorithm with variable learning rate is faster and more accurate than the momentum BP algorithm.