A soft sensor for carbon content of spent catalyst in a continuous eforming plant using LSSVM-GA

Continuous catalytic reforming (CCR) is an important process in hydrocarbon processing to convert low-octane gasoline blending components to high-octane components for use in high-performance gasoline fuels or as source of aromatics. Carbon deposition rate is a critical performance factor of reforming catalyst and carbon content of spent catalyst would directly influence the subsequent catalyst regeneration; thus it is imperative to monitor the carbon content of spent catalyst in real time. In this paper a soft sensor is proposed using least squares support vector machine (LSSVM) with genetic algorithm (GA) to solve the industrial problem for online estimating the carbon content of spent catalyst in an existing CCR plant, wherein the GA is used to select the free parameters of the LSSVM model. The LSSVM with traditional grid algorithm and artificial neural network (ANN) are also applied to model two soft sensors using the same data sets for comparison. The simulation results show that GA shows outstanding performance than traditional grid algorithm for selecting free parameters of LSSVM; the proposed LSSVM-GA soft sensor can achieve smallest errors and shortest computing comparing with LSSVM and ANN. Then the proposed soft sensor is applied to the existing CCR plant; the predictive values are satisfactory.