Share

Online system identification of a compressor test stand with echo state networks.

Number: 1366

Author(s) : LUDWIG G. H., SCHWEDERSKY B. B., FLESCH R. C. C.

Summary

Compressor testing typically requires specialized test stands with several controlled process variables. The use of advanced control methods, which are often model-based, requires a suitable model, either for its synthesis or for its operation. As most compressor test stand dynamics are moderately nonlinear and time-varying, online estimation of nonlinear models is a valuable strategy because it enables the use of advanced model-based controllers that can accurately capture nonlinearities and adapt to changing dynamics. One type of model that has been explored in the literature for online nonlinear estimation is the echo state network, a recurrent artificial neural network architecture that can be trained efficiently. This paper presents an approach which can be used online to identify a nonlinear model of a compressor test stand based on an echo state network, with its output layer weights being estimated using a recursive least squares algorithm. This scheme enables the identification of a nonlinear model without the computational cost typically required to train recurrent neural networks, making it possible to estimate these parameters in an online manner. The evaluation was performed using a real test stand in a scenario in which a multiple-input and multiple-output model was obtained. An adaptive linear model was used as baseline method. The results show that the proposed approach provides significantly better results than the linear baseline, as the linear model was unable to adequately model the nonlinear characteristics of the test stand. The proposed model presented a 34% lower root mean squared error than the baseline.

Available documents

Format PDF

Pages: 10

Available

Free

Download

Details

  • Original title: Online system identification of a compressor test stand with echo state networks.
  • Record ID : 30033608
  • Languages: English
  • Subject: Technology
  • Source: 2024 Purdue Conferences. 27th International Compressor Engineering Conference at Purdue.
  • Publication date: 2024/07/18

Links


See other articles from the proceedings (80)
See the conference proceedings

Indexing