基于大数据的铜板带成品率预测

Prediction on Copper Strip Yield based on Big Data

针对江西某铜板带生产企业,根据历史数据分析了该企业车间部门铜板带生产表面缺陷,构建了铜板带工艺参数采集大数据平台,并对数据做了清洗和存储,利用Spark实现PCA主成分机器算法在线降维,为了预测车间产品成品率,建立了BP_AdaBoost模型,并在构建好的铜板带生产数据云服务存储仓库中挑选出8个变量作为输入,1个变量作为输出,对铜板带生产成品率进行预测。通过网络训练后,AdaBoost算法改进得到的BP_AdaBoost模型模型成品率预测误差低于0.3,相比于未改进的BPNN模型精确率更高,综合拟合度大于0.95,拟合性能好,在铜板带生产企业实际预测精剪成品率时具有显著的非线性映射关系。

For a copper strip production enterprise in Jiangxi Province,the paper uses historical data to analyze the surface defect of copper strip in production,establishes a big data platform for copper strip process parameter acquisition,which can make data cleaning and storage,thus realizing online dimensionality reduction of the PCA principal component machine algorithm by using Spark.BP_AdaBoost model is established to predict the yield of products.8 variables as input and 1 variable as output are selected from the established data cloud service storage to predict the yield of copper strip production.The results show through network training that the yield error of BP_AdaBoost model is predicted to be lower than 0.3%after improvement of AdaBoost algorithm,which has higher accuracy than the unmodified BPNN model.The comprehensive fitting degree is greater than 0.95,showing good fitting performance.For copper strip production enterprises,it has a significant nonlinear mapping relationship in the actual prediction of yield,which is of practical significance for the prediction of copper strip yield in the production.