基于RBF神经网络法的Zr-4合金管材酸洗工艺模型

Pickling Process Model of Zr-4 Alloy Tube Based on RBF Network

研究了Zr-4合金管材酸洗处理过程中,酸洗去除量、酸水转换时间、冲水时间及酸洗次数对管材氟残留量的影响,并基于径向基(RBF)人工神经网络法建立了Zr-4合金管材酸洗工艺与氟残留的神经网络模型。结果表明:冲水酸水转换时间和冲水时间对氟残留量均有影响,且酸水转换时间的影响更为显著;氟残留量与酸洗次数无明显对应关系。Zr-4合金酸洗工艺的RBF神经网络模型结构为3-5-1,实际值与模拟值的相对误差为9.2%。该神经网络模型具有较高的可靠性,可为Zr-4合金酸洗工艺参数的优化提供参考。

In this paper, the effects of pickling removing （PR）, the time of the pickling to washing （TPW）, the washing Function time（WT） and the pickling times（PT） on fluorine present of Zr-4 alloy were studied, and the Radial Basis Network （RBF） Model was proposed to predict pickling process model of the Zr-4 alloy tube. The results show that fluorine present value is influenced by TPW and WT, and the effect of TPW is more apparent than WT, but the PT has little impact on fluorine present value. The optimal RBF network architecture of Zr-4 alloy tube pickling process is considered to be 3-5-1, and the mean squared error （MSE） is 9. 2%. It is a highly reliable model, and can be used for the optimization of the pickling parameters.