Researches Of Neural Network Predictive Control For Dissolved Oxygen Concentration Based On LM Algorithm

Activated sludge process is an effective method of sewage pollutants which use the degradation effect of microbial in sewage sludge to clear pollutant in wastewater, and which is currently one of the most widely used wastewater treatment processes in engineering. The dissolved oxygen(DO) concentration is one of the extremely significant parameters in this wastewater treatment process. However, owing to the time-varying and nonlinear characters, the DO concentration is difficult to achieve desired control effect with traditional control methods(such as conventional PID control, etc.) in the presence of strong disturbance, parameters uncertainties and other circumstances. Neural network predictive control(NNPC) fully takes advantage of its nonlinear mapping ability of neural network on model predictive. And combining mechanism of predictive control feedback correction, rolling optimization, NNPC is more suitable for the control of such nonlinear systems.Inspired by the above discussions, in this topic, a NNPC method based on Levenberg- Marquardt(LM) algorithm for DO concentration is proposed from model predictive, feedback correction rolling optimization three aspects. The main work is given as follows:Firstly, a simplified DO model is established after reasonable hypotheses and constrains in terms of activated sludge model No.1(ASM1) proposed by International Water Association(IWA). And the simplified DO model reflects the intrinsic relationship among the concentration of dissolved oxygen, activated sludge concentration, and substrate concentration.Secondly, aiming at the drawbacks of falling into local minimum easily and slow convergence speed in BP neural network, the Levenberg-Marquardt algorithm(LM algorithm) has been used to improve the general BP neural network. Furthermore, model identification performance of LM-BP neural network has been verified through simulation.Thirdly, aiming at the issues in the control of DO concentration, a NNPC method is proposed. The control system is designed from model predictive, feedback correction and rolling optimization.At last, the effectiveness of the LM-BP neural network for the prediction model of DO concentration has been demonstrated through simulation. By comparing with traditional control strategies, the dissolved oxygen setpoint tracking performance and robustness has been confirmed.The innovation and contribution of this paper are as follows:Firstly, by employing the LM algorithm, the BP neural network is optimized, which overcomes the shortcomings of traditional neural network and improves the accuracy of model predictive in NNPC for DO concentration.Secondly, considering the difficulty of the DO concentration in tracking control, a NNPC method has been proposed. And the controller has been designed from model predictive, feedback correction, rolling optimization and other aspects. What's more, compared to conventional PID control and model predictive control(MPC), the simulation results show that NNPC achieves better adaptability and robustness and brings obvious improvement in tracking performance.