Enhancing Soft Sensor Performance in Control Loops: Integrating State-Dependent Parameter Method with Regression-based Dynamic Data Reconciliation Filter.

This study aims to design a soft sensor using the state-dependent parameter (SDP) method within the control loop. Soft sensors estimate variables based on measured values of other variables, but measurement noise affects their performance. The advantage of the SDP method is its reliance on the system state for model parameter dependency, leading to a lower model order and a reduction in number of input variables. However, this introduces measurement noise into the soft sensor outputs. The research focuses on developing a reliable soft sensor that effectively handles measurement noise, improving accuracy and reliability within the control loop. The innovation of this paper lies in enhancing the SDP soft sensor (SDPSS) using the dynamic data reconciliation (DDR) filter. The proposed approach’s superiority stems from utilizing linear regression models instead of complex dynamic process models as redundant models in the DDR filter’s development. The performance of the SDP method is evaluated by employing an actual industrial sulfur recovery unit as a reference point for assessing the effectiveness of soft sensors. Quantitative results conclusively demonstrate that SDP-based soft sensors significantly enhance estimation accuracy compared to alternative soft sensing techniques. A simulated continuous stirred-tank reactor is also used to evaluate the effectiveness of the proposed approach in two scenarios: external disturbance elimination and set-point tracking. The DDR-SDPSS’s performance is compared with that of the statistical filter-based SDPSS, demonstrating improved variable monitoring, reduced susceptibility to fluctuations, and satisfactory performance across a wide range of measurement noise variances without the need for controller readjustment.