Real time identification for the hydrogen anaerobic production process based on multiple differential neural networks and takagi-sugeno fuzzy decision method

In nature, many systems show very complex behaviors. Among others characteristics, some of those plants exhibit a high degree of oscillations throughout the time or metastable locations. In particular, biological systems which are slow compared with physical or chemical ones, includes lots of examples where those metastable regions are really important to understand their dynamics. Moreover, those metastable states produce very complex mathematical models that usually are inaccuracy or fail to reproduce such real behaviors. In real life, if the mathematical model has the aforementioned failures, the process control may become into a hard task. The anaerobic digestion satisfies all aforementioned characteristics. Indeed, when this process is used to produce methane or hydrogen in continuous regimen, many metastable regions will appear. On the other hand, to obtain an accuracy model of such process may be simplified using mass balance methods. Nevertheless, no one model could reproduce the trajectories observed in real digestion process. A natural alternative to solve this problem is to use adaptive algorithms to obtain approximation models. Nevertheless, adaptive algorithms used to approximate such difficult behaviors can also show important deficiencies. Many adaptive non-parametric algorithms could not reconstruct the trajectories of such complex dynamics. The differential neural network (DNN) is not an exception. Indeed, when just one DNN is applied to achieve the approximation, the modeling error may be not so close to zero. One possible suggestion to solve this problem is to construct a set of DNN working in parallel. The members of such set will work each one on well-defined trajectories subspaces where the uncertain system evolves. How to combine the identification properties offered by the DNN and the characteristic decision capabilities provided by fuzzy methods is the main topic discussed in this chapter. The selection of which neural network is activated depends on decision achieved by a Takagi-Sugeno fuzzy system. Even when the proposed method works efficiently in numerical simulations, the proposed method should demonstrate its workability in real situations. That is why we reported the construction of low-cost embedded device that implements the designed adaptive modeling scheme. The designed device works with similar quality to that performed in numerical results.