Adaptive recurrent neural network control of biological wastewater treatment: Research Articles
International Journal of Intelligent Systems - Soft Computing for Modeling, Simulation, and Control of Nonlinear Dynamical Systems
Recursive Bayesian recurrent neural networks for time-series modeling
IEEE Transactions on Neural Networks
Dynamic Neural-Network-Based Model-Predictive Control of an Industrial Baker's Yeast Drying Process
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
Wastewater treatment process (WWTP) is difficult to be controlled because of the complex dynamic behavior. In this paper, a multi-variable control system based on recurrent neural network (RNN) is proposed for controlling the dissolved oxygen (DO) concentration, nitrate nitrogen (SNO) concentration and mixed liquor suspended solids (MLSS) concentration in a WWTP. The proposed RNN can be self-adaptive to achieve control accuracy, hence the RNN-based controller is applied to the Benchmark Simulation Model No.1 (BSM1) WWTP to maintain the DO, SNO and MLSS concentrations in the expected value. The simulation results show that the proposed controller provides process control effectively. The performance, compared with PID and BP neural network, indicates that this control strategy yields the most accurate for DO, SNO, and MLSS concentrations and has lower integral of the absolute error (IAE), integral of the square error (ISE) and mean square error (MSE).