| 贝叶斯框架下的自适应质量变量预测模型 |
| |
| 引用本文： | 朱雨婷,田 颖.贝叶斯框架下的自适应质量变量预测模型[J].教育技术导刊,2021,20(1):103-108. |
| |
| 作者姓名： | 朱雨婷 田 颖 |
| |
| 作者单位： | 1. 上海理工大学 机械工程学院；2. 上海理工大学 光电信息与计算机工程学院，上海 200093 |
| |
| 基金项目： | 国家自然科学基金青年项目（61903251） |
| |
| 摘 要： | 针对工业生产过程中的时变性问题，提出贝叶斯网络框架下的自适应质量变量预测建模方法。采用改进的即时学习策略，将数据库分成若干局部数据子集，快速选择与待测样本相似度较高的一组数据作为训练样本， 再利用主成分分析对训练样本过程变量进行特征提取，借此作为网络模型输入变量。利用基于改进Figueiredo-Jain算法的EM算法估计高斯混合模型参数，构建高斯混合模型逼近贝叶斯网络联合概率密度，训练得到贝叶斯网络下的自适应质量变量预测模型。基于田纳西伊斯曼（TE）仿真过程获得的数据，利用该方法对成分XG进行预测并与传统PCA-BN模型对比。结果证实该方法最大误差下降14.4%，均方根误差下降7.5%，相对误差下降8.3%，验证了该方法解决时变性问题的有效性。
|
| 关 键 词： | 贝叶斯网络 即时学习 EM算法 高斯混合模型 质量变量预测 |
| 收稿时间： | 2020-10-23 |
An Adaptive Quality Variable Prediction Model in a Bayesian Framework |
| |
| ZHU Yu-ting,TIAN Ying.An Adaptive Quality Variable Prediction Model in a Bayesian Framework[J].Introduction of Educational Technology,2021,20(1):103-108. |
| |
| Authors: | ZHU Yu-ting TIAN Ying |
| |
| Institution: | 1. School of Mechanical Engineering， University of Shanghai for Science and Technology；2. School of Optical-Electrical and Computer Engineering， University of Shanghai for Science and Technology， Shanghai 200093， China |
| |
| Abstract: | To solve the problem of time variability in industrial processes， a modeling method of adaptive quality variable prediction in the framework of Bayesian networks is proposed in this paper. The improved just-in-time learning adaptive strategy is adopted to divide the database into several local data subsets and quickly select a group with high similarity to the samples to be tested as training samples. The principal component analysis is used to extract the characteristics of the process variables of the training samples， which can be used as the input variables of the network model， not only eliminating the correlation between variables， but also simplifying the network structure. Then， the EM algorithm based on the improved Figueiredo-Jain algorithm is used to estimate the parameters of the Gaussian mixture model， and the established Gaussian mixture model is used to approximate the joint probability density in the Bayesian network， and the adaptive quality variable prediction model under the Bayesian network is trained.Based on data obtained from Tennessee Eastman （TE） simulation process， the prediction results of component XG by this method is compared with the traditional PCA-BN model，and the results show that the maximum error is decreased by 14.4%， the root mean square error is decreased by 7.5%， and the relative error is decreased by 8.3%， which verifies the effectiveness of this method in solving the time-variation problem. |
| |
| Keywords: | Bayesian networks just-in-time learning EM algorithm Gaussian mixture model quality variable prediction |
|
| 点击此处可从《教育技术导刊》浏览原始摘要信息 |
| 点击此处可从《教育技术导刊》下载免费的PDF全文 |
