Multi-model adaptive decoupled estimation of component concentration for nonlinear convection-diffusion process based on information fusion

• A novel multi-model decoupled estimation method is proposed for nonlinear processes. • Real-time estimation is achieved by fusing sensor measurements with mold predictions. • The inherent decoupled feature overcomes limitations of coupled state-space methods. • The method's advantages are shown through comparison with the extended Kalman filter. Online monitoring of component concentration in convection-diffusion processes is in high practical demand. Variations in fluid flow rates cause significant nonlinearity in the input (source terms) to output (concentration fields) mapping. Additionally, online concentration measurements often contain substantial noise. This paper presents a novel method for online concentration estimation in nonlinear convection-diffusion processes. First, the process is divided into linearized subspaces based on inlet flow velocity. Then, step response prediction sub-models are constructed based on the process mechanism, and a scheduling strategy for these sub-models is established. Finally, guided by the multi-model adaptive concept, a multi-model adaptive prediction model for convection-diffusion process component concentration is obtained. Combined with a linear decoupled estimation scheme, adaptive decoupled estimation is achieved by fusing measured and predicted temporal information. Based on the aforementioned approach, adaptive decoupled estimation of component concentrations is achieved in a two-dimensional convection-diffusion system under time-varying flow velocities. The proposed method is compared with the extended Kalman filter, demonstrating its effectiveness in overcoming the challenges faced by existing state-space model-based information fusion techniques.