CAJ | 학술논문

为了提高极限学习机对化工过程的高维数据进行建模的能力，提出了一种基于信息熵优化的学习算法。利用互信息方法判断输入变量与输出变量之间的相关性，通过去除部分与输出变量相关性较弱的输入变量来过滤冗余信息，从而达到降维的目的。然后利用熵权法对输入数据进行加权优化，从而降低输入数据中的离散点对极限学习机模型精确度的影响。因此本文提出了一种基于信息熵的ELM算法。该算法以UCI标准数据集进行测试，并以PTA工业系统数据进行实际验证。实验结果表明，与传统 ELM 算法相比，优化后的学习算法在处理高维数据时具有稳定性强、建模精度高的特点。从而拓展了神经网络技术在化工领域里的应用。
위료제고겁한학습궤대화공과정적고유수거진행건모적능력，제출료일충기우신식적우화적학습산법。이용호신식방법판단수입변량여수출변량지간적상관성，통과거제부분여수출변량상관성교약적수입변량래과려용여신식，종이체도강유적목적。연후이용적권법대수입수거진행가권우화，종이강저수입수거중적리산점대겁한학습궤모형정학도적영향。인차본문제출료일충기우신식적적ELM산법。해산법이UCI표준수거집진행측시，병이PTA공업계통수거진행실제험증。실험결과표명，여전통 ELM 산법상비，우화후적학습산법재처리고유수거시구유은정성강、건모정도고적특점。종이탁전료신경망락기술재화공영역리적응용。
The extreme learning machine (ELM) is not very effective for high dimensional data modeling. To solve this problem, the mutual information was adopted. The mutual information was used to determine the correlation between the input and output variables, and then the input variables that had weak correlation were filtered. As a result, the dimension reduction was carried out. And the weight optimization of input data was realized by means of the entropy method. Then the impact of the discrete points in the input data was reduced. Thus, an optimization ELM algorithm based on an information entropy method was proposed. To verify the proposed model, the UCI standard data sets, at the same time, the actual industrial data of PTA were selected. The results show that, the proposed model performs better in the treatment of high dimensional data with than traditional ELM does. The proposed model has better stability and higher modeling accuracy. Thereby, the application of neural network technology in the chemical industry was expanded to some extent.