Development of soft sensor system via dynamic optimization

In chemical process industry, some variables are difficult to measure on-line due to the limitation of measurement techniques, reliability or high cost to install a hard sensor. Some measurable variables, such as product quality, cannot be used for real-time optimization and control due to the large time delay. The soft-sensing of these variables is considered as an efficient method, and remains as an open problem. A new soft sensor system is developed to solve these problems in this work, which is based on rigorous model by using dynamic optimization to minimize the bias square between the model outputs and the measured outputs. Compared with other soft sensor systems, the proposed soft sensor system accurately employs the nonlinear model and considers the constraints in the optimization. It not only is more efficient in economic but also can be used on-line for real-time optimisation and control. The soft sensor system is successfully applied to estimation of the feed composition of a pilot heat-integrated distillation column system.