一种PCA优化的自编码模型降阶方法  

A reduced order method based on PCA optimization and autoencoder model

作　　者：王杰峰 姜超颖 卫薇 纪元 WANG Jiefeng;JIANG Chaoying;WEI Wei;JI Yuan(Information Center,Guizhou Power Grid Co.,Ltd.,Guiyang 550000,China;Library,Xidian University,Xi’an 710071,China)

机构地区：[1]贵州电网有限责任公司信息中心,贵州贵阳550000 [2]西安电子科技大学图书馆,陕西西安710071

出　　处：《西安邮电大学学报》2023年第5期86-91,共6页Journal of Xi’an University of Posts and Telecommunications

基　　金：陕西省自然科学基础研究计划项目(2022JM-336);南方电网有限责任公司科技项目(GZKJXM20200770)。

摘　　要：对数字孪生中的高维度模型的降阶问题进行研究,提出一种主成分分析(Principal Component Analysis,PCA)优化的自编码模型降阶方法(PCA and Autoencoder based Reduced Order Method,PCA-Auto-ROM)。该方法通过分析PCA和自编码器各自的特性构造高维数据的降阶模型,利用全连接层对降维后的数据进行分类。为了验证所提降阶方法的有效性,构造仿真数据集,将其与基于多层自编码器的降阶方法(Autoencoder based Reduced Order Method,Auto-ROM)、基于PCA的降阶方法(PCA based Reduced Order Method,PCA-ROM)进行对比。仿真结果表明,对于存在线性相关的高维数据,所提降阶方法能够有效减少原始数据中的冗余信息,平均分类准确率高于Auto-ROM和PCA-ROM,可以提高数据分类的准确度。在同等数据信息损失程度下,所提降阶方法能够减少训练过程的计算代价。In addressing the dimensionality reduction challenge of high-dimensional models in digital twin,a principal component analysis(PCA)and autoencoder based reduced order method(PCA-Auto-ROM)is proposed.The dimensionality reduction models for the high-dimensional data are constructed by analyzing the individual characteristics of the PCA and the autoencoder,and the reduced-dimensional data are classified by the fully connected layers.To validate the effectiveness of the proposed order reduction method,simulated datasets are built and compared with the autoencoder-based reduced order method(Auto-ROM)and the PCA-based reduced order method(PCA-ROM).Simulation results show that,for high-dimensional data with linear correlations,the proposed method can effectively reduce the redundant information in the original data,and can improve the accuracy of data classification with the average classification accuracy increased compared to the Auto-ROM and the PCA-ROM.At an equivalent level of data information loss,the proposed method can reduce the computation cost during the training process.

关 键 词：数字孪生 主成分分析 自编码器 降阶方法 

分 类 号：TP14[自动化与计算机技术—控制理论与控制工程]