A New Neural Observer for an Anaerobic Bioreactor

Abstract

In this paper, a recurrent high order neural observer (RHONO) for anaerobic processes is proposed. The main objective is to estimate variables of methanogenesis: biomass, substrate and inorganic carbon in a completely stirred tank reactor (CSTR). The recurrent high order neural network (RHONN) structure is based on the hyperbolic tangent as activation function. The learning algorithm is based on an extended Kalman filter (EKF). The applicability of the proposed scheme is illustrated via simulation. A validation using real data from a lab scale process is included. Thus, this observer can be successfully implemented for control purposes.

Dynamic learning rate (&#x03B7;<inf>D</inf>) for recurrent high order neural observer (RHONO): Anaerobic process application

“…Since the pH measure is directly related to IC, this variable is estimated with a negligible error during all the simulation. In general, pH D γ η = these estimation errors could be due to the observer structure, which is a simple one [1]. The proposed η D with initial nominal values produces a diminution in the transient error as compared with Fig 3. Fig.…”

“…2. First, η D with the nominal values of η determined in [1], second, η D values 100 % larger than nominal ones and third, η D values 50 % smaller than nominal ones. Learning rate initial values are selected in heuristic way as is discussed in [24].…”

“…2 displays the disturbance in S 2in at t=200 h. Fig. 3 shows that good estimations for X 2 , S 2 and IC are obtained with the RHONO proposed in [1], with an error when the step on the input substrate is incepted, which is eliminated in the steady state. As was mentioned previously, this error could be due to the observer structure.…”

“…Nominal values of η were determined in [1] for each one of the estimated variables as: η 1 for the estimation of X 2 , η 2 for the estimation of S 2 and η 3 for the estimation of IC. In order to test the observer sensitivity to input changes, a disturbance on the input substrate (270% S 2in increase) is incepted at t = 200 h, as shown in Fig.…”

“…In a previous publication [1], the activation function of the proposed observer scheme is defined as a hyperbolic tangent one and the learning rate (η) of the EKF training algorithm was set to a nominal value. The observer scheme was tested in presence of disturbances and parameters variation in order to verify its performance.…”