Scholarlink

In the present work, two generic identification systems are designed and implemented, applicable to industrial processes through artificial neural networks of NARX time series and later a simplified version of the same, GADALINE. The algorithm is developed online and offline, providing clear and precise information about the behavior and critical parameters of the system to be identified. For the execution and validation of this new technique, first, measurements are made within a real plant to measure liquid levels. The new identification methods are compared with another conventional system identification technique, LAPLACE, thus revealing that these methods have various applicable structures and configurations with excellent potential for predicting response signals. The results showed that the complete NARX neural network and a simplified GADALINE neural network improve the identification of a system compared to the identification through LAPLACE, verified by implementing a controller tuned based on the identification of a system through these techniques.

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

Identification of Liquid-Level Systems in Industrial Processes Through Artificial Neural Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jorge Goiburú,
  • Jorge Vera,
  • Enrique Fernández Mareco,
  • Diego P. Pinto-Roa

摘要

In the present work, two generic identification systems are designed and implemented, applicable to industrial processes through artificial neural networks of NARX time series and later a simplified version of the same, GADALINE. The algorithm is developed online and offline, providing clear and precise information about the behavior and critical parameters of the system to be identified. For the execution and validation of this new technique, first, measurements are made within a real plant to measure liquid levels. The new identification methods are compared with another conventional system identification technique, LAPLACE, thus revealing that these methods have various applicable structures and configurations with excellent potential for predicting response signals. The results showed that the complete NARX neural network and a simplified GADALINE neural network improve the identification of a system compared to the identification through LAPLACE, verified by implementing a controller tuned based on the identification of a system through these techniques.