Stable neural-adaptive control of activated sludge bioreactors

This paper proposes an adaptive neural network control for an activated sludge bioreactor used for waste-water treatment. The novel method prevents weight drift and associated bursting when a persistent disturbance affects the system, without sacrificing performance - unlike traditional e-modification. The neural adaptive method outperforms two types of PI controllers when tracking arbitrary set points, of organic substrate and dissolved oxygen, when appropriate feedforward terms are unknown. The method also outperforms a feedback linearizing controller using model parameter estimates when an observer is used to provide an estimate of unmeasured substrate concentration.