Soft Sensor Development for Nonlinear and Non-Gaussian Processes in Different Noisy Conditions based on Statistical Perspectives

Soft sensors are employed in industrial processes to estimate the variables that required a sophisticated and complex procedure to be measured. However, in many practical cases, the processes possess a nonlinear and multimodal characteristic, which may pose some obstacles in the estimation. Although various modeling approaches have been considered in this regard, the problem remains an interesting subject both in academia and in industry. This paper proposes a novel soft sensor structure that can address both nonlinearity and non-Gaussianity, along with considering the environmental noise effects. The aforementioned soft sensor uses statistical regression models such as Gaussian Process Regression (GPR) with the consideration of mixture models. We applied the proposed soft sensor to a well-known benchmark. The simulated results illustrate promising performances for even moderate noise intensities. Furthermore, a comparison with some previous studies is provided.