Hybrid modular mechanistic/ANN modelling of a wastewater phosphorus removal process

The hybrid modular mechanistic/ANN technique was applied to a wastewater phosphorus removal process with simulation data. The hybrid model was able to describe accurately the process dynamics. With the expectation-maximisation identification technique, the mixture of experts (ME) network converged to a configuration where the two experts developed independent expertise in describing either the phosphorus release state or the phosphorus storage state, which is precisely the main metabolic switch occurring in this process. The final ME network provided a more structured representation of the underlying cellular kinetics. This resulted in higher accuracy and generalisation capacity of the final hybrid mechanistic/ME model.