Reseach On Neural Network Based Real Time Optimization Control For Wastewater Treatment Processes

Wastewater treatment processes (WWTPs) are unknown large nonlinear systemssubjected to significant perturbations in flow and load, together with variation in thecomposition of the incoming wastewater, therefore, the stable manipulate of WWTPscontrol is the most significant problem. Meanwhile, the heavy electric energyconsumption and manipulation cost of wastewater treatment plants in our country arewidespread, energy reducing under the constraint of wastewater treatment efficiencyis another important problem.Real time optimization (RTO) control method is becoming an effective methodfor complex process industry optimizing and controlling. Real time optimizationcontrol is combined with close loop control and process manipulation optimization, itcan be divided into two layers. The upper layer calculated the optimized controlvariables set points through the optimization of economic performance index; thelower layer tracking control the optimized set points, to make the control processmanipulated in an economic optimized state. For WWTPs, the upper layer aims atminimizing the energy consumption under the constraint of effluent quality, calculatesthe optimimal control variable set points; the lower layer tracking control theoptimized set points, aims at the stable manipulation of WWTPs and higher controlprecision.Artificial neural network (ANN) is an evaluate tool for realize RTO control.ANN can be a good online controller because of its excellent online learning ability;and the approximate ability of ANN makes it has good performance in nonlinearsystem modeling. Augmented Lagrange multiplier (ALM) method is suitable fornonlinear programming problem, and it is also valuable for RTO upper layer’soptimization problem.Focusing at the real time optimization control problem of WWTPs, the mainresearch works of this paper are listed as follows:(1) WWTPs characters analysis and Benchmark set upThis paper focus at pre-denitrification wastewater treatment processes, analyzedthe WWTPs characters and the influent characters under three weather conditions.International Water Association (IWA) and European cooperation COST announcedand international simulation benchmark of WWTPs: Benchmark Simulation ModelNo.1(BSM1). This paper analyzed the characters of biological reactor and secondarysettler, and tested the effectiveness of BSM1.(2) A neural network online modeling and controlling method is proposed The research objectivs of tracking control layer are the insurance of the stable ofsystem manipulation and higher control presicion. This paper proposed a neuralnetwork online modeling and controlling (NNOMC) method for nonlinear, strongdecoupling, time-varying and model-unknown WWTPs. For the time-varying andmodel-unknown characters, a modeling neural network (MNN) is proposed for onlinemodeling the WWTPs via its approximate ability; and for the nonlinear and strongdecoupling characters of WWTPs, proposed an neural network controller (NNC) foronline tracking control performance via its online learning ability. The stability ofNNOMC control WWTPs system is also discussed, and it is clear that the systemstability can be assurance by the proper choosing of hidden layer learning rates ofboth NNC and MNN. The proposed NNOMC is applied in the simulation experimentof dissolved oxygen (DO) concentration control and WWTPs multi-variables control,the results show NNOMC method can insurance the system manipulation stability,and it has higher control presicion and multivariable decoupling ability.(3) An neural network based augmented Lagrange multiplier optimizationmethod is proposedThe optimization problem of WWTPs can be divided into two parts: the effluentquality restriction achieving and the system energy reducing. The WWTPsconstrained optimization problem is designed via the above two parts. The WWTPsconstrained optimization problem is a black-box optimization problem because of theunknown relationship among energy consumption, effluent quality and controlvariables set points of WWTPs. Aiming at this problem, this paper proposed a neuralnetwork based augmented Lagrange multiplier (NNALM) optimization method.Online modeling the energy consumption function and effluent quality function forthe balck-box model, and then online optimizing the WWTPs constrainedoptimization problem. The NNALM algotithm stability can be assuranced by theproper choosing of the learning rates of the augmented Lagrange method viaLyapunov theorem. Applied NNALM method into the real time optimization ofWWTPs constrained optimization problem, the simulation results show NNALMmethod can calculates the control variables set points effectively via the constrainedoptimization problem and system information feedbacks.(4)The RTO control of WWTPs based on NNOMC and NNALMCombined the proposed NNOMC method and NNALM method, this paperproposed a real time optimization control structure of WWTPs. The parameters ofNNOMC method, NNALM method and the optimization range of control variablesset points is properly chosed in simulation experiment set up. This paper simulated theNNALM and NNOMC based RTO control method for WWTPs under three weatherconditions (dry weather, rain weather and storm weather), and the results indicatedthat the WWTPs system can be manipulated stable under the NNOMC control which achieves better control presicion; and the system energy is effectively reduced underthe constraint of effluent quality via the optimization of NNALM.