基于肖维勒准则与主元分析的有机朗肯循环神经网络建模方法  

Modeling Method of Organic Rankine Cycle Neural Network based on Chauvenet Criterion and Principal Components Analysis

作　　者：平旭 杨富斌[1] 张红光[1] 邢程达 杨海龙 王焱[1] PING Xu;YANG Fubin;ZHANG Hongguang;XING Chengda;YANG Hailong;WANG Yan(Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学,北京100124

出　　处：《大电机技术》2023年第6期70-76,共7页Large Electric Machine and Hydraulic Turbine

基　　金：国家自然科学基金(51906119);北京市自然科学基金(3222024);天津大学内燃机燃烧学国家重点实验室2020年开放课题(K2020-08)。

摘　　要：随着人工神经网络技术的不断发展,有机朗肯循环(organic Rankine cycle,ORC)神经网络模型广泛应用于系统分析和优化领域。针对现有ORC神经网络模型计算量大、时间周期长和精度偏低的问题,本文提出了基于肖维勒准则与主元分析的ORC神经网络建模方法。采用肖维勒准则对ORC实验数据进行预处理,以去除异常数据,同时数据得到规范化处理。随后,采用主元分析对ORC特征进行矩阵变换和降维,以提取与ORC运行显著相关的特征向量。最后,通过实验数据验证了提出方法的有效性。该方法可在提高模型精度的同时,降低建模所需的时间。与基于原始数据的ORC神经网络模型相比,基于该方法的ORC神经网络模型建模所需时间降低了88.69%。同时,模型预测精度提高了19.93%。With the development of artificial neural network technology,the organic Rankine cycle(ORC)neural network model is widely used in the field of system analysis and optimization.In response to the ORC neural network models with large calculations,long time cycles and low accuracy,this paper proposes the ORC neural network modeling method based on the Chauvenet criterion and principal components analysis.The ORC experimental data are preprocessed by using the Chauvenet criterion to remove the abnormal data,and the data are normalized.Subsequently,the principal components analysis is used to transform and diminish the matrix of ORC features to extract significantly related feature vectors with ORC operation.Finally,the effectiveness of the proposed method is verified by experimental data.This method can reduce the time required for modeling while improving the model accuracy.Compared with the original data-based ORC neural network model,the time to model the ORC neural network model modeling based on this method is reduced by 88.69%.At the same time,the model prediction accuracy is increased by 19.93%.

关 键 词：有机朗肯循环 人工神经网络建模 肖维勒准则 主元分析 

分 类 号：TK172[动力工程及工程热物理—热能工程]