Abstract
This paper presents an extension of adaptive learning rate (ALR) scheme utilized for single output networks to the case of multiple outputs regarding Lyapunov stability. The proposed ALR is employed for training a particular type of polynomial nonlinear state space (PNLSS) identification model in generalized predictive control (GPC) of a system with general multicompartment lung mechanics. Recursive system identification is performed in the output error prediction context where the observed input-output data are processed sequentially. Grey box recursive system identification is performed. Training of the PNLSS model is performed using gradient descent with ALR that guarantees convergence. The same approach of ALR also provides controller stability. Simulations are performed assuming the system dynamics are unknown and compliance parameters of the system vary over time as well as an external disturbance affects the system. Results demonstrate that, the proposed ALR improves robustness of the adopted control strategy even at high level of disturbance and parametric unceartinty. It proves to be a reliable tool for stability of feedforward network training where it is compared to some of well-known adaptive gradient descent studies in the literature.
Keywords
Subject Areas
Related YTU Activities
Sustainability activities and projects connected to this article.
