Study On The Multi-objective Optimum Based On Multi-intelligence-algorithm And Application In Anaerobic Ammonium Oxidation System

Anaerobic ammonium oxidation?Anammox?has been regarded as an efficient process to treat nitrogen-containing wastewater.However,the treatment process is not fully understood in terms of reaction mechanisms,process simulation and control.The parameters such as pH,temperature and the substrate concentration of ammonia,nitrite can produce non-ignorable impacts of the reaction.In the meantime,with increasingly stringent regulations of effluent quality,process monitoring and control have become essential in wastewater treatment.The research efforts on numerical modeling and performance optimization of these techniques have also been limited because of their multiphysics nature and the complexity of synergistic effects.To overcome these problems,software sensors in combination with intelligent controls were developed.In this paper,the multi-objective optimization problem existing in anammox was analyzed after a comprehensive understanding of the status of the application of anammox for nitrogenous wastewater.Based on some pioneering work of intelligent algorithms used in wastewater treatment,and basic structure of principal component analysis?PCA?,back-propagation neural network?BP?,a least square support vector machine?LSSVM?,and non-dominated sorting genetic algorithm-II?NSGA-II?,a multi-objective control strategy mixed soft-sensing model is developed to systematically design the operating variations for multi-objective control by integrating the developed predictive model and optimization model.The main contents and conclusions in this research are described as follows:?1?A up-flow anaerobic sludge blanket reactor inoculated with anaerobic sludge from a wastewater treatment plant in Guangzhou was adopted to investigate the start-up of ananmmox via increasing the concentration of influent substrate and gradually shortening hydraulic retention time?HRT?in the reactor.The process was successfully start-up within 86days'operation,and the removal loading rate of TN was 0.53 kg N?m^-3?d^-1.During the reactor stable operation phase,the stoichiometry molar ratios of the NO₂^--N conversion and NO₃^--N production to NH₄⁺-N conversion were calculated as?1.25±0.03?:1 and?0.28±0.02?:1respectively.The removal efficiency of ammonium and nitrite were 99%and 95%with an influent total nitrogen?TN?loading rate of 0.90 kg N?m^-3?d^-1 at 6 h HRT,which denoted that the reactor can maintain stable and efficient nitrogen removal performance.?2?Glucose was added as OM disturbance after the process start-up,the operating efficiency of anammox under different nitrogen load and COD interference was explored by changing the influent conditions.The experimental results showed that the reactor had a preferably not only nitrogen removal,while the removal efficiency of COD was better in a low C/N ratio?<0.50?when the influent ammonia nitrogen and nitrite concentration were about 200mg/L and 264 mg/L,respectively.On the other hand,the input parameters including all initial concentrations and pH had significant effects on the effluent ammonia nitrogen and total nitrogen removal amount.It also can be observed that the derivatives of effluent profiles can accurately detect the ends of the effluent ammonia nitrogen and total nitrogen removal concentration,and be a useful information source.?3?A new hybrid model for anammox parameters forecasting was introduced by integrating BP,LSSVM and PCA in this paper.A different number of sample points selected parameters would affect the polyhydric relationship between auxiliary variable and sample points,thereby affecting the accuracy of the prediction model in using PCA for dimensionality reduction process.The actual and forecast results showed that the PCA-BP and PCA-LSSVM models for predicting the effluent water quality followed the dynamics in anammox system well.In contrast,the PCA-LSSVM model had better prediction performance,in that the mean absolute percentage error and root mean square normalized error were smaller,and its model had a higher correlation coefficient between the predicted value and the true value.?4?A multi-objective control strategy mixed soft-sensing model based on PCA-LSSVM approach and NSGA-II was proposed to intelligently control the process and optimize the treatment system in anammox process.Based on results obtained,the Pareto frontier gave us the reasonable solutions of pH,NH₄⁺-N,NO₂^--N and COD level in influent,to achieve a better removal effect,the influent pH should be kept between 7.50 and 7.52,the COD/TN ratio is suggested to maintain at 0.15 and the NH₄⁺-N/NO₂^--N ratio is suggested to maintain at 0.62.The experimental results showed that the effluent established on the basis of the optimization results was close to the actual real value.As a result,the proposed model is effective and feasible,which can provide reference and guidance for the design and operation of anammox process.