Separable least squares identification of nonlinear Hammerstein models: application to stretch reflex dynamics

Area of Science:

• * Systems biology and biomedical engineering
• * Nonlinear systems analysis and modeling

Background:

• * Hammerstein cascades, comprising a zero-memory nonlinearity and a linear filter, are valuable for modeling complex biological systems.
• * Accurate parameter estimation for Hammerstein cascades is challenging, with current iterative methods having limitations.

Purpose of the Study:

• * To develop and evaluate a novel separable least squares (SLS) optimization method for simultaneous estimation of Hammerstein cascade parameters.
• * To compare the performance of the SLS method against traditional iterative algorithms in modeling biological systems.

Main Methods:

• * Development of a separable least squares algorithm tailored for Hammerstein cascade identification.
• * Application of the algorithm to electromyogram (EMG) data from stretch reflex experiments.
• * Monte-Carlo simulations to assess model robustness under various input signal conditions.

Main Results:

• * The proposed SLS algorithm yielded more accurate models, demonstrating superior prediction of system responses to novel inputs compared to the traditional iterative method.
• * SLS approach effectively identified Hammerstein cascade parameters without bias, even with non-Gaussian, nonwhite input signals.
• * Traditional iterative methods showed biased model estimations under similar non-ideal experimental conditions.

Conclusions:

• * Separable least squares optimization provides a more robust and accurate method for identifying Hammerstein cascade models in biological systems.
• * The developed SLS algorithm overcomes limitations of existing iterative techniques, particularly with complex experimental data.