A multi-operating LPV system modelling method based on a high-order ARX model

This paper initially proposes an identification method based on high-order ARX models. This method employs the AIC to search for the optimal high-order structure and applies a suboptimal order reduction algorithm to the high-order model to maximize the fit index. The proposed method enhances model identification accuracy under colored noise conditions. Subsequently, for multi-operating nonlinear systems, a LPV system is adopted, constructed from linearly weighted multiple local models and weighting functions. This includes the design of scheduling variables and weighting functions. Innovatively, a gradient descent approach is introduced to update the weighting functions, alongside the design of the objective function and convergence conditions for the algorithm. This methodology optimizes the global LPV system design to improve model accuracy. The proposed method is validated through a simulation case study, where it is compared against an approach based on the IV method. Under NSR of 1 %, 5 %, and 10 %, the proposed method demonstrates superior parameter identification accuracy for local models and better fit results for the LPV model compared to the IV approach. Finally, the method is applied to model an industrial powder metallurgy sintering furnace. The experimental results confirm the feasibility and effectiveness of the proposed approach.