ANN modelling of pyrolysis utilising the characterisation of atmospheric gas oil based on incomplete data

Processing of atmospheric gas oils (AGOs) in the petroleum industry by pyrolysis is an important task. In order to improve control of production and to predict the pyrolysis product yields, it is necessary to use novel approaches. The combination of selected analytical as well as modelling methods described in this contribution seems to be very promising and usable also for other types of pyrolysis feedstocks. Employment of an artificial neural network (ANN) model as the prediction tool is the crucial point. The main problem is the treatment of complex mixtures. While analytical methods used here give the picture of global characteristics and group composition, the method used to characterise the complex mixture by a well-defined substitute mixture of real components provides the input information for the ANN model. Unfortunately, the required measured data, typically distillation and other curves, are often incomplete and it is impossible to obtain directly the ‘phase portraits’ needed to establish the substitute mixture. Possibilities of how to solve this particular problem utilising other information about the mixture, e.g. bulk properties, are shown and applied to concrete AGO feedstocks with satisfactory results.