摘要
基于梯度增强回归树(GBRT)的方法建立打浆度预测模型。采集实际工业环境中磨浆过程变量(如流量,纸浆浓度和磨浆机功率)和原料性质,包括原料纤维形态和浆料性质作为模型输入,所有输入变量数据来源于造纸厂。在实时数据上检验模型精度,均方误差为RMSE^k=0.9948。对比支持向量机(SVM)打浆度模型,GBRT打浆度模型时间复杂度更低。
A soft sensing method for beating degree modeling method is proposed based on the gradient boosting regression tree(GBRT) algorithm. In the model structure process, the refining process variables(including flow rate, pulp concentration, and refiner power) and the raw material properties(including fiber morphology and stock properties) are selected as model input. All the data of these input parameters are collected from a real-world paper mill. The model accuracy is tested by the real-time data, the mean square error of the soft sensing results is RMSE^k = 0.9948. Compared with the support vector machine(SVM) model, the proposed GBRT model has lower time complexity.
作者
孟子薇
洪蒙纳
李继庚
满奕
MENG Ziwei;HONG Mengna;LI Jigeng;MAN Yi(State Key Laboratory of Pulp and Paper Engineering, South China University of Technology,Guanghzou 510640,China)
出处
《造纸科学与技术》
2019年第1期83-88,共6页
Paper Science & Technology
关键词
磨浆
打浆度
软测量
梯度增强回归树
pulping process
beating degree
soft sensing technology
gradient boosting regression tree