System Identification and Control of a Biotrickling Filter
Abstract This paper studies empirical modeling and control of a biotrickling filter (BTF) used for air pollution control. Step response transfer function (TF) with first-order-plus-time-delay model and steady-state artificial neural network (NN) model were developed for BTF based on input–output (I/O) data obtained from simulation of a rigorous model. These simple models offer fast predictions compared to the rigorous model and render control implementation for BTF feasible. Gas velocity and inlet concentration of hydrogen sulfide (H2S) (target pollutant) were considered as the main process inputs while outlet concentration of H2S was selected as the BTF performance variable (output). The TF and NN models fitted well with the I/O data and the resulting regression coefficient values were above 0.97. Different simulations with the fitted NN model were performed and compared with the rigorous model data at steady state. The NN model perfectly captured the steady-state behavior of the BTF process. Two control strategies were implemented, namely proportional–integral/feedback control and model predictive control, also known as receding-horizon control. The controllers were based on the fitted TF model representation of BTF under study. For the control structure, gas velocity, inlet concentration, and outlet concentration were selected as manipulated, disturbance and controlled variables, respectively. Through set-point and disturbance change tests, it was observed that the model predictive controller offered superior set-point tracking capabilities while the feedback controller showed better control in dealing with disturbances. However, both controllers provided adequate control in general.