Scholarlink

<p>In industrial processes, it is highly challenging to simultaneously identify both the system structure and parameters under finite and low-quality measurements. In this paper, a transfer identification method based on sparse Bayesian learning is proposed for the nonlinear autoregressive with exogenous input model, by incorporating knowledge from a similar but distinct system to improve identification accuracy and reduce identification costs. To effectively leverage useful knowledge from the source system, a structure-transfer step and an observation-boosting predictor are designed based on the sparse Bayesian learning method. Furthermore, we introduce the Kullback-Leibler divergence as a unified similarity metric to quantify the discrepancy between the transfer model and the ideal model. By minimizing this divergence, an optimal transfer estimation for the target parameter vector is derived, ensuring consistency and informativeness of the transferred knowledge. Finally, a numerical example and a continuous stirred tank reactor example are simulated to demonstrate the superiority of the proposed method.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Transfer identification for NARX models based on sparse Bayesian learning

  • 期刊论文

  • 发表日期：

    2026年3月11日

查看全文
  • Siyuan Li,
  • Shuang Gao,
  • Xiaojing Ping,
  • Feng Ding,
  • Xiaoli Luan

摘要

In industrial processes, it is highly challenging to simultaneously identify both the system structure and parameters under finite and low-quality measurements. In this paper, a transfer identification method based on sparse Bayesian learning is proposed for the nonlinear autoregressive with exogenous input model, by incorporating knowledge from a similar but distinct system to improve identification accuracy and reduce identification costs. To effectively leverage useful knowledge from the source system, a structure-transfer step and an observation-boosting predictor are designed based on the sparse Bayesian learning method. Furthermore, we introduce the Kullback-Leibler divergence as a unified similarity metric to quantify the discrepancy between the transfer model and the ideal model. By minimizing this divergence, an optimal transfer estimation for the target parameter vector is derived, ensuring consistency and informativeness of the transferred knowledge. Finally, a numerical example and a continuous stirred tank reactor example are simulated to demonstrate the superiority of the proposed method.