A novel neuro Takagi Sugeno modeling approach for an activated sludge reactor
Matoug Lamia, Khadir M. Tarek · 2016
This paper investigates the use of Artificial Neural Networks (ANN), more precisely Multi-Layer Perceptrons (MLP), as an alternative to the weighting function μ in a Multi-Model approach type Takagi Sugeno (TS) for an Activated Sludge Reactor. The reduced bio-reactor activated sludge ASM1 model, which describes the biological degradation of an activated sludge reactor, is designed based on several simplifications, as a TS fuzzy model, which structure is based on a set of linear sub models, covering the process input-output space, interpolated by a nonlinear weighting function. Using data gathered from the set of sub model outputs used as inputs of a nonlinear multi-step ANN predictor. The function μ, earlier obtained using a Quasi-LPV approach, is then approximated using the designed MLP where the training and validation of the latter is performed and results are compared with the original approach with and without input and parametric disturbances.