Gaussian process and its application to soft-sensor modeling

With the estimation of key quality index in an industrial naphthalene distillation column,a novel soft-sensor modeling method based on Gaussian process(GP)was proposed for complex industrial processes.The principle of automatic relevance determination,implemented with GP model,was proposed to determine the secondary variables for the soft-sensor.To overcome the shortcomings existing in present methods,which can not determine the measurement uncertainty of soft-sensors,the GP based soft-sensor was developed to get both the prediction of key quality index and its measurement uncertainty simultaneously.Application studies showed that the GP soft sensor model not only determined the secondary variable automatically,but also possessed both high accuracy and small measurement uncertainty,which met the demands for reliable measurements in industrial application.