Neural network and on-off control of bicarbonate alkalinity in a fluidised-bed anaerobic digester

A laboratory-scale fluidised-bed anaerobic digester with a sintered glass carrier, Siran®, was operated for 8 months on a simulated bakerʹs yeast wastewater (12,000 mg soluble COD l−1) at a loading rate of 27 kg COD m−3 d−1, giving 75% removal of soluble COD. Percentage CO2, H2 concentration, gas flow rate and pH were measured continuously. An on-line bicarbonate alkalinity (BA) monitor was used in experiments comparing two control strategies, adjusting digester buffering by addition of NaHCO3 solution during organic overloads. The first, an on-off controller with a set point at the steady-state level (2700 mg CaCO3 l−1), maintained BA concentration but resulted in levels above the upper set point. Thus, to avoid consuming excess NaHCO3 the rate of delivery and solution strength must be carefully adjusted. The second was a controller developed from a neural network trained on BA data from an anaerobic filter operating on ice-cream processing wastewater (alkalinity around 1400 mg CaCO−1). Without re-training, despite the different steady-state BA levels and reactor type, the neural network based controller was capable of maintaining stable BA levels during overload without overshoot. Control of BA during overloads did not prevent changes in gaseous CO2 and H2 concentrations and gas flow rate.