基于RFECV-LightGBM-TPE联合模型的在制品质量问题预测方法研究

Research on Quality Problem Prediction Method of WIP Based on RFECV-LightGBM-TPE Joint Model

为对在制品进行质量预测,提出一种基于RFECV-LightGBM-TPE联合模型的在制品质量问题预测方法。该方法通过RFE递归特征消除和CV交叉验证进行特征筛选,确定特征维度。针对实时采集的在制品工艺参数不充分的情况,通过结合特征重要性的方法改进KNN缺失值填补算法,补全未采集的工艺参数,解决在制品难以进行质量预测的问题。同时利用历史数据训练LightGBM算法构建质量预测模型,并通过TPE算法进行参数寻优提高预测精度。最后以博世生产线采集的工艺参数为例将本文所提模型与常见模型进行对比,结果表明本文所提模型可以更好的对质量问题进行预测,给智能工厂的建设提供了参考价值。

A method of predicting the quality of a product in process based on the RFECV-LightGBM-TPE model is proposed.The feature dimension is determined by RFE recursive feature elimination and CV cross validation.Aiming at the situation of insufficient process parameters collected in real time,the KNN missing value filling algorithm is improved by the method of feature importance to complete the uncollected process parameters,so as to solve the problem of difficult quality prediction.At the same time,the LightCBM algorithm is trained with historical data to build a quality prediction model,and parameter optimization is carried out by the TPE algorithm to improve the prediction accuracy.Finally,taking the process parameters collected from the Bosch production line as an example,the model proposed in this paper is compared with the common models.The results show that the model proposed in this paper can better predict the quality problems and provide reference value for the construction of intelligent factory.