Decentralized indirect adaptive Fuzzy-Neural Multi-Model control of a distributed parameter bioprocess plant

Abstract

The paper proposed to use recurrent Fuzzy-Neural Multi-Model (FNMM) identifier for decentralized identification of a distributed parameter anaerobic wastewater treatment digestion bioprocess, carried out in a fixed bed and a recirculation tank. The distributed parameter analytical model of the digestion bioprocess is reduced to a lumped system using the orthogonal collocation method, applied in three collocation points (plus the recirculation tank), which are used as centers of the membership functions of the fuzzyfied plant output variables with respect to the space variable. The local and global weight parameters and states of the proposed FNMM identifier are implemented by a Hierarchical Fuzzy-Neural Multi-Model Sliding Mode Controller (HFNMM-SMC). The comparative graphical simulation results of the digestion wastewater treatment system identification and control, obtained via learning, exhibited a good convergence, and precise reference tracking outperforming the optimal control.

Distributed Parameter Bioprocess Plant Identification and I-Term Control Using Decentralized Fuzzy-Neural Multi-Models

“…3. The structure of the entire control system, (Baruch et al, 2008a;Baruch et al, 2008b;Baruch et al, 2008c) contained Fuzzyfier, Fuzzy Rule-Based Inference System (FRBIS), and defuzzyfier. The FRBIS contained five identification, five feedback control, five feedforward control, five I-term control, five total control T-S fuzzy rules (see Fig.…”

“…The structure of the entire control system, (Baruch et al, 2008a;Baruch et al, 2008b;Baruch et al, 2008c), contained Fuzzyfier, Fuzzy Rule-Based Inference System, containing twenty T-S fuzzy rules (five identification, five sliding mode control, five I-term control, five total control rules), and a defuzzyfier. Due to the learning abilities of the defuzzifier, the exact form of the control membership functions is not need to be known.…”

“…The models are implemented on FFNN using Backpropagation (BP) and LevenbergMarquardt learning algorithms. Similarly to the static ANNs, the fuzzy models could approximate static nonlinear plants where structural plant information is needed to extract the fuzzy rules, (Baruch et al, 2008a, Baruch et al, 2008b, Baruch et al, 2008cBaruch & GalvanGuerra, 2009). The difference between them is that the ANN models are global models where training is performed on the entire pattern range and the FS models perform a fuzzy blending of local models space based on the partition of the input space.…”

“…The paper of (Baruch et al, 2008a) proposed as a dynamic function in the consequent part of the T-S rules to use a Recurrent Neural Network Model (RNNM). Some results of this RNNM approach for centralized and decentralized identification of dynamic plants with distributed parameters are given in (Baruch et al, 2008a;Baruch et al, 2008b;Baruch et al, 2008c;Baruch & Galvan-Guerra, 2009). The difference between the used in the other papers fuzzy neural model and the approach used in (Baruch et al, 2008a) is that the other one used the Frasconi, Gori and Soda RNN model, which is sequential one, and in (Baruch et al, 2008a), it is used the RTNN model, which is completely parallel one.…”

“…The reduced plant model (58)- (66), could be used as unknown plant model which generate input/output process data for decentralized adaptive FNMM control system design, based on the concepts, given in (Baruch et al, 2008a;Baruch et al, 2008b;Baruch et al, 2008c;Baruch et al, 2008d). The mentioned concepts could be applied for this DPS fuzzyfying the space variable z, which represented the height of the fixed bed.…”