Artificial Neural Network Based Software Sensor for Yeast Biomass Concentration during Industrial Production

The artificial neural network is a potential ´sensor´ in the complex bioprocess. The recurrent neural network (RNN) was employed as the software sensor to measure the biomass concentration during the baker´s yeast industrial production, owing to its good ability in dealing with non-linear and time-varying process. Based on the data sets provided by the plant, input variables were selected as air flow rate (G), ethanol concentration (Eth), volume of the contents in the reactor (Vol), temperature (T), pH and their time-delay values as well as the predicted values of yeast biomass concentration at delayed time. The topology of the RNN was optimized to be 11-16-1. The RNN showed good generalization ability for the testing samples. The robustness of the RNN was evaluated by adding deliberately inflicted noises to the G and Eth. The RNN showed higher robustness to the noise from Eth than that from G