基于AANN特征缩放的虚拟样本生成方法及其过程建模应用  

Novel virtual sample generation based on feature scaling of auto-associative neural network and its applications to process modeling

作　　者：朱宝[1,2] 乔俊飞 ZHU Bao;QIAO Junfei(Information Department,Beijing University of Technology,Beijing 100023,China;Powerchina Resources Ltd.,Beijing 100048,China)

机构地区：[1]北京工业大学,北京市100023 [2]中国电建集团海外投资有限公司,北京市100048

出　　处：《计算机与应用化学》2019年第4期304-307,共4页Computers and Applied Chemistry

基　　金：国家自然科学基金重大项目(61890930)

摘　　要：工业大数据时代的到来推动着智能数据挖掘领域的发展。然而,大数据中的小样本问题严重影响了数据驱动建模的精度。为了解决这一问题,本文提出一种基于自联想神经网络特征缩放的虚拟样本生成方法(FSAANN-VSG)。首先,从自联想神经网络(AANN)的特征层出发,扩缩变换小样本自联想网络模型的特征层信息,从而产生新的特征信息,随后经前向计算得到虚拟样本;所提方法采用AANN模型,一方面能够生成符合原始小样本知识的虚拟样本,另一方面能够去除样本间的噪声信息;最终实现样本量增加,同时有助提高模型的精度。为验证本文方法的有效性,首先采用UCI数据库中的Concrete Slump Test (CST)数据集,随后将所提的方法应用于乙烯生产过程建模,仿真结果验证了本文所提方法的有效性,加入虚拟样本后,模型的精度更高、鲁棒性更好。The arrival of the industrial big data era has promoted the development of data mining. However, the small sample problem in big data seriously affects the accuracy of data-driven methods. In order to solve this problem, a virtual sample generation based on feature scaling of auto-associative neural network(FSAANN-VSG) is proposed. In the proposed FSAANN-VSG method, firstly, from the feature layer of the auto-associative neural network(AANN), the feature layer information of auto-associative network model for small samples is scaled to generate new feature information;then virtual samples are obtained using forward calculations. The AANN model was utilized in the proposed FSAANN-VSG method, on the one hand, virtual samples can be generated in the knowledge scope of original small samples;on the other hand, the noise information between samples can be removed;finally, the sample size is increased and the accuracy of model can be improved. In order to verify the effectiveness of the proposed method, a UCI data called Concrete Slump Test(CST) is selected;then the proposed method is applied to modeling the Ethylene production process. Simulation results show that the proposed FSAANN-VSG method can be of aid for small sample modeling with high accuracy and robustness after adding the virtual samples.

关 键 词：虚拟样本生成 自联想神经网络 特征缩放 过程建模 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP389.1[自动化与计算机技术—控制科学与工程]