Researches On Identification Modeling For Dissolved Oxygen And NARMA-L2 Control Based On Neural Network

As a classical wastewater treatment measure,the activated sludge process is extensively applied in urban wasterwater treatment.Nevertheless,the activated sludge process involves numerous reaction parameters and at the same time wastewater treatment process is a nonlinear and long time delay process,so it is often a great challenge to achieve the desired control performance in practice.On account of the traditional PID control process cannot solve the control problem of nonlinear system with variable parameters.Therefore,the Bayesian regularization optimized neural network is applied to identify the nonlinear dissolved oxygen concentration model.Afterwards,neural network control system is established,for the sake of controlling the concentration of dissolved oxygen by feedback linearization.The main points of operation are as follows:In the first place,this paper describes the traditional wastewater treatment process,the activated sludge process and the corresponding reaction rate equation in detail.The widely used the activated sludge model No.1 is introduced.For the sake of facilitating the subsequent model identification and neural network control process,the activated sludge model is reasonably simplified and the stability of the simplified system equation is analyzed.In the second place,considering that the simplified system is a time-varying nonlinear system that changes with time,the neural network is applied to identify the established nonlinear system.The advantages and disadvantages of various neural network identification structures are analyzed and the parallel structure is selected as the identification structure.Bayesian regularization is applied to optimize BP neural network to perfect the accuracy of identification model,for sake of ameliorating the ordinary weaknesses of BP neural network.For extruding the superiority of BR-BP identification model,the nonlinear model is selected and the improved elephant herding algorithm is applied to optimize the BP neural network and the system identification is compared with the neural network optimized by scaled conjugate gradient method.In the end,the controllability of the simplified system is analyzed and NARMAL2 neural network control system is designed to control the sewage treatment system.NARMA-L2 neural network control,sliding mode control and PID tracking control results are compared by simulation and the control performance comparison results of three control methods under external interference are given.The results verify the accuracy and anti-interference of NARMA-L2 control system on the concentration of dissolved oxygen.The innovation and contribution of this paper are as follows:1.Bayesian regularization is used as learning algorithm to optimize neural network.It overcomes the shortcomings of neural network and improves the accuracy of neural network identification model in NARMA-L2 control.2.NARMA-L2 neural network control method for dissolved oxygen concentration is designed to achieve better tracking control effect and anti-interference ability than conventional PID and sliding mode control.