Soft sensors for on-line biomass measurements.

One of the difficulties encountered in control and optimisation of bioprocesses is the lack of reliable on-line sensors for their key state variables. This paper investigates the suitability of using on-line recurrent neural networks to predict biomass concentrations. Input variables of the proposed recurrent neural network are feed rate, liquid volume and dissolved oxygen. Experimental results revealed that the proposed neural network is able to predict biomass concentrations with an accuracy of +/-11%.