Data-Driven Soft Sensors Based on Support Vector Regression and Gray Wolf Optimizer

Data-driven soft sensors offer a reliable means of estimating hard-to-measure variables using easily measurable process variables, so they play a crucial role in implementing closed-loop control in industrial processes. These soft sensors enable real-time process control, act as backup measurement systems, facilitate what-if analysis, support sensor validation, and contribute to fault diagnosis. Additionally, they allow for a reduction in the number of required hardware sensors, thereby enhancing system reliability while decreasing costs associated with sensor acquisition and maintenance. The proposed method for soft sensor design involves utilizing the Support Vector Regression (SVR) technique which effectively addressing nonlinear regression problems. To optimize the SVR parameters, the Gray Wolf Optimization algorithm (GWO) is employed. The GWO-SVR model is then used for predicting system variables. To validate the effectiveness of this approach, case studies are conducted in the benchmark Tennessee Eastman process. The results demonstrate the successful application of the proposed method.