Robust Hierarchical Identification of Hammerstein Systems With Latent Time‐Delay

ABSTRACT In order to identify a class of Hammerstein systems characterized by latent time‐delay based on the data containing outliers, a robust identification method called hierarchical re‐weighted least angle regression‐least squares (RW‐LAR‐LS) algorithm is proposed. First, a decomposition based LAR algorithm is proposed to efficiently and accurately determine the time‐delay by introducing the largest input regression length. Then, a matrix whose elements are adaptively adjusted by the designed rule is integrated into the LS algorithm to reduce the impact of outliers, achieving highly accurate parameter estimates. The proposed RW‐LAR‐LS algorithm is efficient and more robust to the outliers and has better parameter and delay estimation performance compared to the over‐parameterized LS algorithm and the LAR‐LS algorithm. A numerical example and a heat exchange system are provided to demonstrate the effectiveness of the proposed algorithm.