基于机器学习的铝电解电容器用阳极铝箔电化学性能预测

Prediction of electrochemistry performance of anodic aluminum foil for aluminum electrolytic capacitor based on machine learning

阳极铝箔的化成工序是一个多影响因素相互作用的材料加工过程,分析各因素的影响规律及交互作用的难度较大。针对以上问题,本文采用随机森林模型计算变量权重系数并进行特征筛选,在此基础上建立基于神经网络的阳极铝箔电化学性能预测模型。结果表明:终端检定电压是对耐压值和比电容影响最大的工艺参数,耐压值和比电容的大小随着终端检定电压的增加分别呈阶梯式增加和阶梯式减小的变化。将经过随机森林算法特征筛选后保留的11个变量作为输入量,分别建立耐压值、比电容神经网络预测模型,经过参数调优后预测模型的R^(2)分数分别为0.987和0.982,电化学性能的预测值与实测值的匹配度较高。通过随机森林模型、神经网络模型的建立和分析,研究阳极铝箔生产过程参数与电化学性能指标之间的定量关系,可实现工艺参数优化与性能预测,以及生产过程重要工艺参数的识别与控制。

The formation of anodic aluminum foil is a kind of material processing process in which many factors interact with each other.It is difficult to analyze the influence mechanism and interaction of each factor.In this paper,the weight coefficient of variables are calculated by random forest model,and the characteristics are selected.The prediction model of electrochemistry performance for anodic aluminum foil based on neural network is established.The results show that,the terminal-verification voltage is the main process parameters which has significant influence on withstand voltage and specific capacitance,which increases and decreases step by step with the increase of terminal verification voltage,respectively.The 11 variables retained after characteristic selection of random forest algorithm are taken as input variables,and BP neural network prediction models of withstand voltage and specific capacitance are established respectively.After parameter optimization,the ^(R2) scores of prediction models are 0.987 and 0.982,respectively,which means the predicted value of electrochemistry performance matches the measured value well.Based on the establishment and analysis of random forest model and neural network model,the quantitative relationship between the process parameters and electrochemistry performance of anodic aluminum foil is studied.Therefore,the parameters optimization and performance prediction,as well as the identification and control of important process parameters during the production process are realized.