引用本文:杨鑫宇,吕政,赵珺,王伟.基于迁移学习的离心式水泵扬程性能预测[J].控制理论与应用,2021,38(5):615~622.[点击复制]
YANG Xin-yu,LV Zheng,ZHAO Jun,WANG Wei.Transfer learning-based performance prediction of centrifugal pumps[J].Control Theory and Technology,2021,38(5):615~622.[点击复制]
基于迁移学习的离心式水泵扬程性能预测
Transfer learning-based performance prediction of centrifugal pumps
摘要点击 1681  全文点击 531  投稿时间：2020-08-14  修订日期：2020-11-30
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2021.00537
  2021,38(5):615-622
中文关键词  离心式水泵  迁移学习  神经网络预测  最小二乘归纳式迁移学习
英文关键词  centrifugal water pump  transfer learning  neural network prediction  least square induction transfer learning
基金项目  国家重点研发计划项目(2017YFA0700300), 国家自然科学基金项目(61522304, 61533005, 61703070)资助.
作者单位E-mail
杨鑫宇 大连理工大学 控制科学与工程学院 yangxinyu1996@mail.dlut.edu.cn 
吕政* 大连理工大学 控制科学与工程学院 lvzheng@dlut.edu.cn 
赵珺 大连理工大学 控制科学与工程学院  
王伟 大连理工大学 控制科学与工程学院  
中文摘要
      离心式水泵作为工业领域常见的抽水机械, 一直有着广泛的应用. 然而在其性能指标预测过程中, 理论 模型难以达到高精度要求, 而机器学习模型难以应用于多工况环境. 本文提出了一个最小二乘归纳式迁移学习 (LSITL)方法, 该方法利用离心式水泵扬程性能曲线特征, 通过最小二乘方式提取迁移知识, 并利用归纳法建立多工 况下的迁移模型, 再通过最小二乘支持向量机(LSSVM)方法的反向求解实现对离心式水泵的性能预测. 本文通过 与机理建模方法和传统机器学习方法的对比, 表明了本文中方法具有准确性高, 适用范围广的优势, 可以实际应用 到离心式水泵性能指标的预测当中.
英文摘要
      Centrifugal water pumps have been widely used as pumping machines in the industrial field. However, in the process of predicting the performance indicators of the pumps, it is always difficult for not only the theoretical model to meet the high-precision requirements, but also for the application of the machine learning model to the multi-condition environment. Therefore, a least square induction transfer learning method (LSITL) is proposed in this paper. The new method uses the characteristics of the centrifugal water pump head performance curve to extract the migration knowledge through the least square method, and use the induction method to establish the migration model under multiple working conditions. And it also realizes the performance prediction of the centrifugal pump by the reverse solution of the least square support vector machine method. By comparing with mechanism modeling methods and traditional machine learning methods, the new method proposed is this paper shows the advantages of high accuracy and wide application range, and can be applied to the prediction of centrifugal pump performance indicators.