论文部分内容阅读
采用改进神经网络,和残液法不必分离同时测定拼混染料各组分的上染率。运用改进的网络预测并计算出染色前后的吸光度差,运用残液法公式算出上染率。针对网络易出现“局部极小”、“收敛速度慢”等原因给神经网络加入动量因子,改进学习率自适应等并探讨隐层选择神经元数。本文波长点的优化采集光谱分析到的数据,从而尽量避免重复以及错误信息对预测结果的影响。预测拼混染料的浓度中,相对误差最大为0.87%,最小为0.07%,满足生产需要。
Using improved neural network, and the raffinate method does not have to be separated at the same time determine the dye-dye dye components. Using improved network prediction and calculation of the absorbance difference before and after staining, the residue method is used to calculate the dye uptake rate. In view of the network prone to “local minima ”, “slow convergence ” and other reasons to add momentum factor to the neural network, to improve the learning rate adaptive, and to explore the number of hidden layer selection neurons. Wavelength optimization of this paper collected the spectral analysis of the data, so as to avoid duplication and error information on the prediction results. Predict the concentration of blends dye, the maximum relative error of 0.87%, a minimum of 0.07%, to meet production needs.