Journal of System Simulation ›› 2025, Vol. 37 ›› Issue (10): 2652-2661.doi: 10.16182/j.issn1004731x.joss.24-0557

• Papers • Previous Articles    

Robust Identification of Dual-rate Sampled Nonlinear Systems Based on SALR Network

Jiang Wenbin1, Cao Yuqing2, Xie Li1, Yang Huizhong1   

  1. 1.School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China
    2.Wuxi ATI Group Technology Co. , Ltd. , Wuxi 214161, China
  • Received:2024-05-23 Revised:2024-07-04 Online:2025-10-20 Published:2025-10-21
  • Contact: Xie Li

Abstract:

A robust identification algorithm based on the self-join adjacent-feedback loop reservoir (SALR) network was proposed for dual-rate sampled nonlinear systems with complex nonlinear characteristics and measurement outputs containing outliers. The SALR network was applied to describe the nonlinear characteristics of the target system, and wavelet neurons were injected into the reservoir to enhance its memory and nonlinear description capabilities. The identification problem of the nonlinear system was transformed into the identification problem of the network's output weight matrix. The Huber loss function was used to construct the criterion function, and an error threshold was introduced to improve the robustness of the stochastic gradient identification algorithm against outliers. To solve the problem of output data loss caused by dual-rate sampling, the concepts of auxiliary model identification and interaction estimation theory were introduced into the recursive identification process of the output weights, where the estimated outputs of the network were used to replace the unmeasured outputs. Moreover, the whale optimization algorithm was adopted to optimize the network's hyperparameters, further enhancing the identification accuracy. Numerical simulation results validate the effectiveness of the proposed algorithm.

Key words: dual-rate sampling, robust identification, nonlinear system, auxiliary model, recurrent neural network, whale optimization algorithm

CLC Number: