| Sewage treatment is a major measure to prevent and control water pollution and protect the water environment.It is also a large-scale and complex system,so the treatment process is characterized by nonlinearity,strong interference,and uncertainty.The activated sludge method is currently the most widely used method in the industrial scene of sewage treatment,but the process of this method involves many parameters and biochemical reactions,making it difficult to meet the important parameters of the effluent water quality.The dissolved oxygen(DO)concentration is the most important indicator in the activated sludge process,which affects the metabolic activities of microorganisms in the reaction process.Therefore,this paper aims to realize the precise control of DO concentration in the process of sewage treatment,and designs two algorithms for the control of DO concentration in different scenarios.The main research work is as follows:Ⅰ.Build a simulation platform based on the activated sludge No.1 modelResearch the process flow of the activated sludge method,and conduct detailed analysis of the biochemical reactions,material components,stoichiometric parameters and kinetic parameters in the activated sludge No.1 model.The benchmark platform(BSM1)proposed based on the activated sludge No.1 model was built on Matlab Simulink,and the data under different weather conditions were input into the model to obtain the change process of dissolved oxygen concentration,which was used for the subsequent Modeling and implementation of the control strategy provided usable data.Ⅱ.Proposed regularized model predictive control(MPC)algorithm for abnormal DO concentration modelAiming at the sudden change of organic matter concentration in water quality when the sewage treatment process is affected by extreme weather,a regularized MPC controller is designed to reduce the impact of abnormal parameters or disturbances on the system.The algorithm applys the linearized model of DO concentration as the prediction model,introduces L2-norm term into the MPC cost function to punish the abnormal sequence,and determines the selection rules of regularization parameters by analyzing the analytical solution of the algorithm.In the simulation stage,the performance of the algorithm under different regularization parameters is evaluated,and the comparison experiment with the classical algorithm verifies that the proposed algorithm has higher control performance.Ⅲ.Establishment of dissolved oxygen concentration prediction model based on q-RBF networkThe neural network is applied to solve the problem that the activated sludge process is complex and difficult to model,and the learning algorithm of the radial basis neural network(RBFNN)is improved by introducing the q-gradient descent method to improve the learning efficiency of the network.In the simulation stage,different q parameter values are compared for training,and it is verified that qRBFNN has a faster convergence speed than the traditional RBF network.The network is trained by the experimental data simulated by the BSM1 platform,and a high-precision DO concentration prediction model is obtained.Ⅳ.Designing q-RBF neural network model predictive controllerAiming at the control problem of DO concentration under three different weather conditions,an online gradient MPC controller is designed.The controller applys the q-RBF dissolved oxygen concentration model as the predictive model,adopts the idea of gradient descent to update the control rate,and obtains the gradient by calculating the Jacobian matrix between the network output and the system input at each moment,thereby improving the iteration of the algorithm speed and reduce the complexity of the algorithm.The simulation tracked and controlled the DO concentration under three weather conditions,and compared with the simulation of the classical algorithm,it was verified that the controller designed in this paper has better control accuracy and anti-interference performance... |
PDF Full Text Request