计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (22): 257-262.DOI: 10.3778/j.issn.1002-8331.2105-0199

• 工程与应用 • 上一篇    下一篇

迁移学习下的极限学习机代理建模方法及应用

黄泽英,李海艳,林景亮   

  1. 广东工业大学 机电工程学院,广州 510006
  • 出版日期:2021-11-15 发布日期:2021-11-16

Surrogate Model Construction Method of Extreme Learning Machine Based on Transfer Learning and Application

HUANG Zeying, LI Haiyan, LIN Jingliang   

  1. School of Electromechanical Engineering, Guangdong University of Technology, Guangzhou 510006, China
  • Online:2021-11-15 Published:2021-11-16

摘要:

针对在工程产品的迭代更新或者相似工程在特定需求下的重新设计优化中,使用少量样本构建高精度代理模型的问题,提出了基于迁移学习的极限学习机代理建模方法。结合相似工程产品的历史累积数据和当前产品上采样的少量真实样本,构建变可信度近似模型;融合近似模型生成的随机样本和当前产品的真实样本,构建基于迁移学习的极限学习机代理模型。所提算法使用数值算例进行了验证,并通过构建叉车臂架液压系统变幅缸最大压力代理模型的工程案例做进一步验证,实验结果表明,使用少量样本时,所提算法能显著提升代理模型的预测精度。

关键词: 迁移学习, 极限学习机, 代理模型, 变可信度

Abstract:

Aiming at solving the problem of using a small number of samples to construct a high-precision surrogate model in the iterative update of engineering products or the redesign and optimization of similar projects under specific requirements, a surrogate modeling method of extreme learning machine based on transfer learning is proposed. Firstly, combine the historical accumulation data of similar engineering products and a small amount of real samples sampled on the current product to construct a multi-fidelity approximate model. Then, combine the random samples generated by the approximate model and the real samples of the current product to construct the surrogate model of extreme learning machine based on transfer learning. The proposed algorithm is verified by a numerical example, and further verified by an engineering case of constructing the maximum pressure surrogate model of the hydraulic system of the boom of a forklift truck. The experimental results show that the proposed algorithm can significantly improve the prediction accuracy of the surrogate model when a small number of samples are used.

Key words: transfer learning, extreme learning machine, surrogate model, multi-fidelity