基于GA-BPNN算法的碳纸原纸性能指标建模预测研究

本研究通过改变碳纤维长度、碳纤维占比、分散剂用量等工艺参数，制备不同碳纸原纸，探究不同工艺参数对其抗张强度、孔隙率、透气性、电阻率的影响，采用遗传算法改进反向传播神经网络算法（GA-BPNN算法），构建了碳纸原纸性能预测模型。结果表明，碳纤维长度与碳纸原纸抗张强度、孔隙率、透气性呈正相关，与电阻率呈负相关；碳纤维占比与碳纸原纸抗张强度呈负相关，与孔隙率、透气性、电阻率呈正相关；分散剂用量与碳纸原纸抗张强度、电阻率呈正相关，与孔隙率、透气性呈负相关；碳纸原纸抗张强度、孔隙率、透气性、电阻率预测模型平均相对误差（MRE）分别为5.49%、5.75%、5.21%、5.54%，预测模型MRE均小于10%，与实验得到的工艺参数对碳纸原纸性能的关系趋势一致。

Study on Modeling and Prediction of Carbon Paper Base Paper Properties Based on GA-BPNN Algorithm

In this study， carbon paper base paper was prepared by changing carbon fiber length， carbon fiber /PVA fiber mass fraction， dispersant dosage and other process parameters.The influences of different process parameters on tensile strength， porosity， air permeability， and resistivity were explored. In addition， genetic algorithm was used to improve the back-propagation neural network algorithm （GA-BPNN algorithm）， to construct a performance prediction model of carbon paper base paper. The results showed that the length of carbon fiber was positively correlated with the tensile strength， porosity， and air permeability of carbon paper base paper， and negatively correlated with the electrical resistivity. The mass fraction of carbon fiber to PVA fiber was negatively correlated with the tensile strength of carbon paper base paper， but positively correlated with porosity， permeability， and resistivity. The dosage of dispersant was positively correlated with the tensile strength and resistivity of carbon paper base paper， but negatively correlated with the porosity and permeability of carbon paper base paper. The mean relative error （MRE） of the prediction models for the tensile strength， porosity， permeability， and resistivity of carbon paper base paper were 5.49%， 5.75%， 5.21%， and 5.54%， respectively， and the MRE of the prediction models were all less than 10%， which was consistent with the relationship trend of the experimental process parameters on the properties of carbon paper base paper.