A self-organizing soft sensor for process control systems: Integrating support vector regression with subtractive clustering

Accurate and timely prediction of critical process variables is crucial for optimizing performance in process control systems. Traditional hardware sensors often face challenges such as high costs, maintenance issues, and response delays. Data-driven soft sensors present a promising alternative by utilizing historical and real-time process data to estimate key variables. This paper introduces an integrated approach that combines support vector regression (SVR) with subtractive clustering and Grey wolf optimizer (GWO) to develop an advanced soft sensor. Subtractive clustering serves as a self-organizing technique, identifying and grouping the most relevant data points from historical data to establish a robust foundation for training the SVR model, thereby capturing complex nonlinear relationships within the process. To further improve the model’s performance, GWO is used to fine-tune SVR hyperparameters, optimizing the search space for the best results. The proposed soft sensor’s prediction accuracy is evaluated using the Tennessee Eastman process benchmark, the Tennessee Eastman process poses significant challenges due to its highly nonlinear nature and the strong interactions among variables, and its uncertainty is analyzed through a quantile regression scheme. Simulation results demonstrate that this approach improves process monitoring, reduces operational costs, and enhances overall system reliability and efficiency.