Wavelet-based neural networks are employed to simplify the model of a distillation column. The idea is to use a simplified hybrid model of the column for the purpose of on-line optimisation of the unit. The hybrid model consists of a neural network part plus a mechanistic model part. This highly simplifies the model while preserving the accuracy of the model together with the availability of the required internal variables of the model. The employed wavelet-based neural networks (i.e. wave-nets) are briefly explained. Then, results of the approximation capability of the wave-net model and the on-line optimisation of the column with the aid of the hybrid model are presented.