高温变换催化剂制备条件的神经网络优化

将影响低汽气比条件下LB型节能高温变换催化剂活性的主要因素作为人工神经网络的特征输入向量,将全部实验数据分为训练集和预测集,运用Matlab神经网络工具箱,按改进的Bayes自动归一化算法建立反向传播神经网络模型,不仅可防止网络陷入局部最小,而且提高了网络训练精度和泛化能力。适当拓宽正交实验各因素的水平范围,经过不同因素、不同水平间的组合模拟,预测出LB型节能高温变换催化剂的最佳制备条件为氧化铈质量分数0.76%、氧化铜质量分数5.8%、氧化铬质量分数8.6%、氧化镧质量分数1.0%、铁液浓度92 g/L、中和过程最终pH值9.5。在最佳条件下试制催化剂在低汽气比下的平均活性达77.6%。

Optimization Based on Neural Network of Preparation Conditions of LB Energy-saving High Temperature Shift Catalyst

The key influential factors on activities of LB energy-saving high temperature shift catalyst were regarded as characteristic input vectors,and activities of catalysts as output vectors.The experimental data were divided into train group and prediction group.Neural network toolbox in MATLAB and Bayesian automated regularization algorithm were utilized to build back-propagation artificial neural network(BPANN)model.BPANN model could not only avoid network sliding into the trap of partial minimum,but also enhance the precision and the generalization of network.The range of levels of orthogonal tests were broadened moderately,and optimal preparation conditions(CeO 0.76%,CuO 5.8%,Cr2O3 8.6%,La2O3 38.6%,iron 92 g/L and final pH 9.5)of LB energy saving high temperature shift catalyst were obtained by simulations.77.6% activity of made catalysts was obtained under this condition.