Thermal Properties Reduced Models by ANN in Process Simulation

Abstract

It’s very time-consuming to evaluate thermal properties by rigorous methods in process real-time simulation, especially when the simulated project relates to multi-units and multi-components, which takes about 70 to 80 percent of the total simulation time. We developed a new reduced method for thermal properties evaluation based on the artificial neural net(ANN), in which we established several reduced evaluation models using ANN, such as models of vapor-liquid equilibrium, models of vapor-liquid enthalpy and models of temperature calculated from given enthalpy. We used the reduced models in a dynamic distillation simulation. Compared with rigorous thermal properties models, the ANN-reduced models could save 10 to 20 times simulation time with a satisfied accuracy. The results show it’s an efficient and effective method.