Multivariable Control Method For Wastewater Treatment Based On Neural Network

Wastewater treatment process, with the characteristics of time-varying, strong coupling, large time-delay, and severe interference, due to the complex biochemical reaction and various operation variables, is difficult to control. Compared to some other countries, our domestic control technology, resulting in comparatively low control effect and from relatively late start. As a consequence, research on the control of wastewater treatment process is very essential and valuable.In this paper, the components, the reaction process, the stoichiometric coefficient, the kinetic parameters and the chatacteristics of the mechanisms model for Activated Sludge Model1(ASM1) are studied. The key parameters, the chatacteristics and the structure of the benchmark simulation model(BSM1) are researched. Meanwhile, the BSM1 platform is established under the MATLAB environment.For conventional PID approach, it is difficult to get a high control accuracy or realize online parameter adjusting. To overcome these disadvantages, based on self-adaptative strategy, a novel PID method is proposed for dissolved oxygen(DO) concentration control during the wastewater treatment process. Due to the universal nonlinear mapping ability and good learning performance of BP neural networks, parameters in the controller are able to adjust adaptively and online. The parameters of neutral network are trained by the learning algorithm, and make sure the real-time and stability of the PID parameters on-line correction. Simulation results demonstrate the effectiveness and accuracy of this adaptive PID controller.To overcome the deficiency of traditional controllers for multivariable nonlinear system in wastewater treatment, the Takagi-Sugeno(T-S) fuzzy neural network is applied to wastewater treatment process. For on the one hand, the multivariable is controlled by the expression of fuzzy knowledge on the basis of neural networks, and on the other, using the learning algorithm can make the parameters adjust online. What’s more, the online tuning of learning rate improves the performance of T-S fuzzy neural networks and adaptivity of controller. The accuracy, stability and effectiveness of T-S fuzzy neural network controller is verified in the experiments for DO and nitrate nitrogen(NO) concentration control.Because of the serious disturbance of wastewater treatment, a self-organized T-S fuzzy neural network based on clustering algorithm is proposed to design wastewater treatment controller in this paper. Initially, there is no fuzzy rule in the controller. Then, with the growing of the self-organized rules which are based on the clustering algorithm, the network can learn and remember the current knowledge, and adjust its structure adaptively. Except for the compactness of network structure, the algorithm also improve the controller performance. After the construction of structure, the network is trained by BP algorithm. Finally, the designed T-S fuzzy neural network is utilized to control the DO and NO concentration in the wastewater treatment process, and simulations results under different environmental conditions show the great effectiveness, good stability, and high accuracy of the self-organized T-S fuzzy neural network controller.