Modelling and control of submerged membrane bioreactor filtration process using neural network model predictive control

Abstract

Membrane bioreactor employs an efficient filtration technology for solid and liquid separation in wastewater treatment process. Development of membrane filtration model is significant as this model can be used to predict fouling development which is later utilized in control development. Most of the available models only suitable for monitoring purpose, which are too complex, required many variables and not suitable for control system design. Artificial neural network (ANN) is a simple and efficient method in modelling of filtration process. In this thesis, the dynamic ANN is used to model the filtration process using the developed submerged membrane bioreactor (SMBR) pilot plant. The accuracy of the dynamic neural network is further improved using the proposed optimization algorithms. These algorithms are developed based on the hybrid particle swarm optimization and gravitational search algorithm (PSOGSA) using cooperative approach. The first cooperative PSOGSA (CPSOGSA-1) is developed using master-slave cooperative technique where one master group and a few slave groups are created. The second cooperative PSOGSA (CPSOGSA-2) is where multiple groups are created, and the best solution found by one of the group will share with other groups. The model performances of the ANN training and testing are assessed using mean square error, mean absolute deviation and correlation coefficient. To establish the model training performance, another set of input output data from heating process is performed. Furthermore, the training performance of the algorithms is tested to minimize ten mathematical functions. The simulation results indicate the proposed algorithms outperformed the existing PSO, GSA and PSOGSA algorithms for the SMBR model. Similar trends of results can be observed for heating process model and for all benchmark functions tested. An improved SMBR trained model is then used for neural network model predictive control (NNMPC) design for permeate flux control as to prevent flux decline in the membrane filtration cycle due to fouling problem. The PSO, CPSOGSA-1 and CPSOGSA-2 algorithms are utilized in NNMPC real-time optimization cost function. From the experimental result, the best filtration control is given by NNMPC with CPSOGSA-2 algorithm. The superiority of the NNMPC in membrane filtration control resulted from real time implementation showed an improvement of 100% overshoot, 7.06% settling time and 11.96% of integral absolute error when compared to PID-PSO.