| Activated sludge process invented in the early 20th century,has become the key technology of municipal wastewater treatment plants(WWTPs).Despite the development of A2/O,SBR and other derivative processes,there are still some problems,such as unstable effluent,difficulty of total nitrogen(TN)removal,low-level automation and so on.According to the investigation,dissolved oxygen(DO)is the major monitoring and control index of biological unit of WWTPs,and most of researchers generally believe DO is significantly related to the pollutant removal performance.However,in practical engineering,low accuracy of monitoring and regulation severely restricts the control efficiency of wastewater biological treatment process,leading to poor removal performance of pollutants.Therefore,an efficient and applicable intelligent control technology for biological units of WWTPs is needed for achieving economic,efficient and stable operation.Aiming at the issues such as difficulty in DO control,low TN removal rate and unstable effluent in biological units,a simulation model of WWTPs was carried out by using feedforward neural network,which was fed with anoxic/aerobic residence time ratio and superficial gas velocity as input parameters,and its feasibility is analyzed.At the same time,the optimal operating conditions were obtained by enlarging the range of parameters and optimizing the neural network algorithm.The model had selective capacity and the ability to analyze actual operation data of biological unit of WWTPs,proving that it has the potential of energy saving and efficiency improving.The main results are as follows:1.Based on the good correlation between DO and anoxic/aerobic residence time ratio and superficial gas velocity in SBR,the effects of anoxic/aerobic residence time ratio and superficial gas velocity as control variables on the denitrification performance of the system were studied.The results showed that the removal efficiency of TN increased from 50.1±5.5%to 64.0±1.2%as anoxic time increased from 0 min to 60 min in the four-hour SBR system,but when the anoxic time was longer than 150 min,the removal efficiency of NH4+-N was only 73.9%,and the removal rate of TN decreased to 56.3%.At the same time,when the superficial gas velocity decreased from 2.0 cm/s to 1.0 cm/s,the TN removal rate increased from 64.0±1.7%to 68.1 ± 1.9%.However,when the superficial gas velocity was lower than 0.6 cm/s,the DO of the system was no more than 2.0 mg/L,causing the inhibition of sludge activity.Based on 124 sets of valid experimental data,a feedforward neural network model with 6-8-1 structure is constructed.Its goodness of fit is 0.9197 and 0.8634 on the training set and complete set respectively.It is proved to be feasible that using the practical engineering controllable parameters(anoxic/aerobic residence time ratio,superficial gas velocity,etc.)as input of the feedforward neural network to simulate the biological reaction process.2.102 sets of sample data were added as supplementary tests of key parameters such as anoxic/aerobic residence time ratio,superficial gas velocity and water feeding pattern.Three optimization algorithms,Levenberg-Marquardt algorithm,Bayesian regularization and scaled conjugate gradient method were used to optimize the simulation model of biochemical reaction process.The results shown that the scaled conjugate gradient method was the preferred algorithm of simulation model because its advantages of reasonable training steps(156 steps),the shortest training time(<<1s)and highest goodness of fit(0.93).In the process of gradient descent,the scaled conjugate gradient method may avoid over-fitting,slow convergence or oscillation near extremum.Based on the simulation model and the predetermined restrictive conditions,the optimal operation parameters are determined by traversal search:50 minutes of anoxic time,0.6 cm/s of superficial gas velocity and 100%of water inflow before anoxic period.Under these conditions,the removal efficiency of TN in SBR process is as high as 77.3±2.4%,which is significantly better than that in control group(55.2±3.0%),which indicated that the feedforward neural network model trained by scaled conjugate gradient method has the ability to select the optimal parameters of biological reaction process,and can provide a basis for intelligent control of wastewater treatment process.3.By systematically analyzing of 7938 sets of operation data from a WWTP in Zhejiang Province for one year,it is shown that when dealing with the daily organic loading rate(OLR)rising,the increase of fan air flow often occurred within 5-10 hours(7.067 hours for average)after influent COD rush.Since HRT in anaerobic/anoxic section is only 5 hours,it is believed that the 2-hour lag period existed in the operation of biological unit of the WWTP,which has the great influence on the stable operation of biological unit.Moreover,air flow needs to be adjusted frequently in a very short time as well as the control accuracy needs to be improved.To this end,the scaled conjugate gradient method,which was achieved in the previous study,was used to train the feedforward neural network model in order to study the fitting of actual operation data.The results showed that taking COD.TN,pH and influent flow rate as inputs,the scaled conjugate gradient neural network air volume simulation model with goodness of fit of 0.7951(fan start-stop controls anoxic/aerobic residence time ratio,fan air flow controls superficial air velocity)is obtained after normalizing and adjusting the ratio factor of the data.Further optimized by removing the 2-hour lag period,the goodness of fit increased to 0.8638.Based on the optimal selection of fan air flow,the response speed of the system is accelerated and the power cost of the fan was reduced by 9.32%.In conclusion,the scaled conjugate gradient neural network simulation model has high efficiency and applicability for the biological unit of WWTPs,and has the potential of operation optimization and energy saving.Instead of DO,using fan air flow as control object can achieve the intelligent on-line control and stable effluent in municipal WWTPs. |
PDF Full Text Request