NEURAL-NETWORK MODELING FOR PHOTOCHEMICAL PROCESSES

More than 40 years of research work in the domain of photochemical eng ineering has shown that quantitative modelling and solving of the radi ant energy conservation equation coupled to the momentum, mass and hea t balances are very difficult and therefore, even taking into account the most recent developments, of little impact on photochemical reacto r design and process optimization. The kinetics of photochemical proce sses depend on light absorption, and the modelling of photochemical re actors and processes must take into account the spatial distribution o f radiation emitted by a given light source (radiation field) and of t he radiation absorbed. This task has proven to be extremely difficult, even under the most favourable experimental conditions (e.g. sensitiz ed reactions) and simplest reactor geometries. On the other hand, empi rical methods of reactor design and up-scaling do not necessarily lead to optimal results, and modelling by means of artificial neural netwo rks holds the promise of solving problems so far beyond the scope of m ethods using the transport phenomena approach. In this work, neural ne tworks have been applied to the modelling of a homogeneous liquid phas e photochemical system: the photolysis of an aqueous uranyl oxalate so lution. The algorithm used to adjust the weights in neural network app lication was back-propagation. The comparison between the calculated a nd experimental data show good agreement, even when simulation was per formed outside the range of the learning set (extrapolated result).