基于数据的废纸配比预测纸浆白度的研究

Data-driven Approach of Predicting Pulp Brightness Based the Ratio of Waste Papers Used

关键词:  废纸配比  纸浆性能  支持向量（SVM）  BP神经网络  预测模型

Key Words:ratio of waste paper  pulp properties  support vector machine  BP neural networks  prediction model

摘要:针对实际生产过程中根据人工经验配比废纸用量导致纸浆性能与预期差别较大的现状,本研究利用纸厂提供的废纸配比和纸浆性能检测数据,使用BP神经网络和支持向量机（SVM）的建模方法,分别采用全部样本数据和样本平均值数据建立基于废纸配比的纸浆白度预测模型。研究结果表明，在模型预测精度、预测稳定性以及模型训练时间等方面,以样本平均值数据作为建模数据集,使用SVM方法建立的纸浆白度预测模型,具有较好的预测精度(2.42%)和良好的稳定性(0.58%),且模型训练时间短(0.2 s),可以满足实际生产过程的需要。

Abstract:Aiming at the significant difference in expectation of the pulp properties estimated by manual experience based on the ratio of waster papers used, in this paper, utilizing the field data from paper mill, two modeling methods, back propagation (BP) neural networks and support vector machine (SVM), were individually applied to the prediction of pulp brightness based on the ratio of waste papers used. The prediction models were developed using all the data set and the average data set respectively. The results showed that, in terms of the prediction accuracy, prediction stability and training time, the prediction model of pulp brightness by SVM method using the average data set not only had good prediction precision (2.42%) and fairly prediction stability (0.58%), but also the fast training process (0.2 s), which could meet the requirements of the paper mill.