CHEMICAL ENGINEERING TRANSACTIONS  
 

VOL. 62, 2017 

A publication of 

 

The Italian Association 
of Chemical Engineering 
Online at www.aidic.it/cet 

Guest Editors: Fei Song, Haibo Wang, Fang He 
Copyright © 2017, AIDIC Servizi S.r.l. 

ISBN 978-88-95608- 60-0; ISSN 2283-9216 

Research on Dissolved Oxygen Control during Biological 

Sewage Treatment 

Bo Dong, Yi Ge, Hao Zhu, Jie Wang, Yongxiang Zhu* 

Chuzhou Vocational and Technical College, Chouzhou 239000, China 

richard311@foxmail.com 

This paper has put forward a dissolved oxygen control method based on MPC. The method has utilized the 

aforementioned simulation software package for generating large amounts of data, acquired state space 

model of the DO value process through identification, and simultaneously designed an MPC controller whose 

parameters are gradually determined by the common trial-and-error method. In simulation, it firstly carries out 

contrast verification of controller performance under different controller parameters so as to determine better 

controller parameters and lay the foundation for DO value process control. Secondly, it conducts control 

performance research on the activated sludge process by parameters-defined MPC controller and a 

comparison between built-in PI control strategies based on IWA and COST Benchmark. Results show that this 

controller performs better, and the DO value becomes more stable and less undulant. The proposed method 

has two advantages: first, fewer DO activities may make the activated sludge process more stable and reliable 

and thus lead to better processing effects; second, poorer DO fluctuations would exert little load for heating 

blowing machines, which is conducive to its energy-saving operation and consequently will provide conditions 

for low-cost operation of the whole activated sludge process. 

1. Introduction 

Biochemical oxygen demand refers to the amount of dissolved oxygen consumed during organism 

decomposition of microorganisms in surface water, and the standard unit of measurement is mg/L. Generally 

speaking, the process of microorganism decomposition can be divided into two phases: the first phase is the 

process during which the organism is converted to carbon dioxide, ammonia and water; the second phase is 

the so-called nitrifying process during which ammonia is further converted to nitrite and nitrate in the forms of 

nitrosobacteria and nitrifying bacteria (Ternes, 1998). BOD commonly refers to oxygen consumption of a 

biochemical reaction in the first phase. BOD reflects the total amount of organisms which can be decomposed 

by microorganisms in water. Water with less than 1mg/L BOD is considered clean water, and BOD of more 

than 3-4mg/L indicates that the water has been polluted by organisms. However, due to the long measuring 

time needed for BOD and restricted organism activities in sewage with great toxicity, it is difficult to obtain an 

accurate measurement (Ternes et al., 1999; Ternes et al., 1999). 

Chemical oxygen demand refers to the amount of oxidant used by oxidizable matters in water during chemical 

oxidation under specified conditions, and the standard unit of measurement is mg/L. During COD 

measurement, organisms are oxidized into carbon dioxide and water (Castiglioni et al., 2006). The level of 

difficulty of chemical oxidation reaction varies among different organisms in water, and thus the chemical 

oxygen demand can only indicate the total oxygen demand of utilizable matters in water under specific 

conditions (Vieno et al., 2007; Tauxe-Wuersch et al., 2005; Perrone and Amelio, 2006). 

Comparing COD with BOD, measurement of COD is not restricted by water quality and has a relatively short 

measuring time. But COD cannot distinguish an organism that can be biologically oxidized from that which is 

difficult to be biologically oxidized, and it also cannot represent the amount of organisms that can be oxidized 

by microorganism. Furthermore, chemical oxidants cannot oxidize all organic matter, and will oxidize some 

inorganic matter (Lagana et al., 2004; Zorita et al., 2009). Therefore, BOD is appropriately adopted as the 

indicator of the degree of organism pollution; when BOD measurement is restricted by water quality, COD can 

be substituted. 

                                

 
 

 

 
   

                                                  
DOI: 10.3303/CET1762206 

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Please cite this article as: Bo Dong,  Yi Ge,  Hao Zhu,  Jie Wang,  Yongxiang Zhu, 2017, Research on dissolved oxygen control during 
biological sewage treatment, Chemical Engineering Transactions, 62, 1231-1236  DOI:10.3303/CET1762206   

1231



Dissolved oxygen plays an important role during the biological sewage treatment through activated sludge 

(Stasinakis et al., 2008). The stability of dissolved oxygen concentration determines the degree of all 

biochemical reactions in sewage. Without enough DO, aerobic microorganisms can neither survive nor bring 

oxygenolysis into play. However, with extremely high DO concentration, unconsumed DO will reflow to 

hypoxia parts along with reflux inside the activated sludge, and the rate of organism oxidation will increase. 

This leads to a decreasing denitrifying nitrogen-removal process due to the absence of or insufficient carbon 

sources (Wagner and Loy, 2002). Moreover, if DO concentration in the aerobic zone is too low or close to 0, 

facultative bacteria will be transferred to anaerobic respiration, and most aerobic bacterium will basically stop 

breathing, while some aerobic bacterium (mostly filamentous bacterium) may grow well and thus their 

dominant positions in the system will cause sludge expansion. Thus, a suitable DO value must be maintained. 

In other words, during biological sewage treatment, DO value control has become necessary for realizing 

quality standardization of sewage treatment (Postigo et al., 2010). 

Under current actual conditions, DO is still in semi-automatic control or even manual control during most 

sewage treatment processes and has commonly adopted the traditional PID control algorithm with 

unsatisfactory control effects (Yu et al., 2009; Metcalfe et al., 2003). As a result, product quality cannot be 

guaranteed, and the environment has suffered serious pollution while at the same time raw materials are 

being severely wasted. Although control research on such a process globally has achieved some results, 

actual production and application requirements still cannot be met. Therefore, how to apply modern intelligent 

control technology and means to achieve a stable and accurate control of DO value is still an exceedingly 

challenging task. 

As dissolved oxygen control of sewage treatment is a control object with complex characteristics including 

nonlinearity, high time lag and strong interference, it is not easy to achieve satisfactory results (Svenson et al., 

2003). If we adopt intelligent control technology and a high-level automatic control system, the effect and 

efficiency of sewage treatment will be greatly improved, resulting in tremendous social and economic benefits 

will be. Consequently, research conclusions of this paper are of high value in terms of both theory and 

application (Metcalfe et al., 2003; Solé et al., 2000). 

2. Materials and methods 

2.1 Model Predictive Control 

The method of model predictive control (MPC) is a new computer control algorithm comprising three elements 

including model prediction, rolling optimization and feedback correction. Its successful application in some 

complex industrial processes such as oil refining, the chemical industry and electric power has attracted much 

attention to MPC. At present, corresponding theoretical research on MPC is a point of widespread interest in 

the control theory field and has become one of the most representative advanced control strategies in the field 

of industrial process control. 

 

Figure 1:  MPC Control process 

In 1978, the model of predictive heuristic control proposed by Richalet et al. has long been applied to 

predictive control algorithm for the actual industrial process, and its core idea is as follows. On the basis of 

control strategy of online optimization, the current state of the system at each sampling time is taken as the 

initial condition. The dynamic model of the process is utilized, and the system response is predicted within a 

limited time domain. An open-loop optimization problem is solved and a control sequence is obtained 

according to the future performance indicator of this model’s optimization object. Then, the first controlled 

1232



quantity of this control sequence is applied to the controlled object. Because on-line rolling optimization is 

adopted in the predictive control algorithm and the difference between the actual system output and the 

predictive model output is used for feedback and correction during optimization, the predictive model output 

can to a certain degree overcome the predictive model’s influences of deviation and some indeterminate 

interference. Hence, MPC control strategy will be selected as the process control method of DO value in this 

study, and the effectiveness of the control strategy will be verified under the benchmark. 

Figure 1 shows the control process of MPC. In this study, the set value is just that of the DO concentration in 

No.5 biochemical reaction tank (aeration tank). It is generally thought that the DO concentration should be 

maintained at around 2mg/L, and the object output is the DO concentration value in the tank as detected by 

sensors. In this control simulation research, the DO concentration can be acquired through an ideal soft 

sensor model which is adopted to read the corresponding data in the process object. The sensor at the time is 

set as ideal sensor; that is, it boasts characteristics such as no time delay and measurement noise 

interference, and the operating variable is the mass transfer coefficient of dissolved oxygen—KLa. 

Figure 2 shows the control principle diagram of the DO process designed under MPC guidelines. Assuming 

that the temperature remains unchanged during processing, in order to maintain a constant DO value in the 

aeration tank, DO concentration appeared as the same measurement as the ideal sensor placed in the 

aeration tank, and a comparison was made between MPC controller and DO set value. The value of operating 

variable is adjusted by operating variables and used to regulate DO concentration in the tank. Repetition of 

procedures including prediction, optimization and feedback correction in such a process will maintain the DO 

concentration at a certain range of set values and finally achieve the goal of DO process control. 

For the assumed control increment of m steps (current or future), Δu(k), Δu(k+1), , Δu(k+m-1). So the 

predictive output of future p steps is y(k+1|k), y(k+2|k),……, y(k+p|k). However, current or control increment of 

future m step (m<p) is then obtained by minimum value of the following secondary goals through calculation: 

     
     

2 2

, 1 , , 1
1 1

min 1 1
     

 

             
p m

y y

l l
u k u k u k m

l i

y k p r k l u k l             (1)

 

It is also subject to the constraints of the following inequalities: 

 

 

 

, 1, ,

, 0, , 1

, 0, , 1

   

    

       

y y k j y j p

u u k j u j m

u u k j u j m

                                                                                                             (2)

 

In formula (1), Γ𝑙
𝑦
 and Γ𝑙

𝑢 are weight matrices used for penalizing specific variables (y or u) in predictive time 

domain as the future set value vector. Although m steps control increment Δu(k), Δu(k+1), , Δu(k+m-1), will 

be calculated in rolling optimization, only the first control increment will be implemented. Thus, when the next 

sampling interval comes, the control domain will move a step further in rolling optimization. Along with new 

output values collected from the process object and repetition of aforementioned calculation process, the first 

new control increment is implemented again and thus the optimization control of the process object can be 

achieved through such repetition. Predictive outputs of object y(k+1|k), y(k+2|k),……, y(k+p|k) are dependent 

on the actual output of present objects. Assuming that y(k) at this time contains influences of immeasurable 

perturbations and measurable noises, then as a result, in the simulation research, measurable noises have 

been manually added into the object output so as to test the effectiveness and dynamic response capacity of 

the control strategy. Except for operating variable KLa, input variables of the aeration tank are all assumed as 

immeasurable perturbations of the system. 

DO the controller

(MPC)
ASMI

KLa

So,sat

So

 

Figure 2:  DO value of the process control 

1233



2.2 Type font and type size 

By the steady-state simulation carried out through the aforementioned BSM1 platform, steady-state simulation 

data can be acquired in terms of various degrees of aeration and thus the continuous time state space model 

can be established in the following form: 

 

 

dx
Ax Bu

dt

y Cx Du

                                                                                                                                                      (3)

 

In the formula, x is the state vector, while u and y are respectively input and output vectors; A, B, C and D are 

respectively state space coefficient matrix. Figure 3 is the step response curve of the model identification by 

the system under various degrees of aeration. The figure has established step response curves directing at 3 

dissolved oxygen concentration grades in the aeration tank respectively at 2mg/L, 1.4mg/L and 0.9mg/L. It is 

generally recognized that DO concentration should maintain at around 2mg/L, and therefore from the step 

response (blue dotted lines in Figure 3) of its identification model, it can be obtained that: 

2.3 Type font and type size 

Simulation research of controller responsiveness is processed through continuous state space coefficient 

matrix with the above method, and controller parameters are set as follows: sampling time Δt=2.5×10-4day 

=20S, Γu=0.01, m=1, p=10. The results are shown as green dashed lines in Figure 4a. In order to verify the 

control performance of the controller, DO set value was changed from 2mg/L to 2.3mg/L when t=0.03day 

during simulation, and the DO concentration in water entry was reduced by 1mg/L when t=0.07day, as shown 

in blue solid lines in Figure 4a. In both cases, the controller can make a quick control response for DO 

concentration and thus achieve better results. 

Meanwhile, in order to verify the control performance of the controller under different parameter settings, 

controller parameters are adjusted as follows: in Figure 4A, the red dotted line is the controller response curve 

obtained by decreasing predicted domain (p=6), and the bluish green dashed line is the controller response 

curve obtained by enlarging the weight value of variable penalty(Γu=0.1). 

-0.010 -0.005 0.000 0.005 0.010 0.015 0.020 0.025 0.030

0.000

0.005

0.010

0.015

0.020

0.025

D
O

 i
n

 m
g

/L

Time in days

 0.9mg/L

 1.5mg/L

 2.0mg/L

 

Figure 3:  Identification model under various degrees of aeration step response curve 

0.00 0.02 0.04 0.06 0.08 0.10

2.0

2.1

2.2

2.3

2.4

D
O

 i
n

 m
g

/L

Time in days     
0.00 0.02 0.04 0.06 0.08 0.10

140

150

160

170

180

190

200

k
la

 i
n

/d
a
y

Time in days  

(a)                                                                                  (b) 

Figure 4:  Controller response performance simulation under various control parameters 

1234



It can be seen from the comparison results in Figure 4a that narrowing predictive domain can shorten 

controller response time, although an overshoot may increase larger; while enlarged input of variable penalty 

will enhance response time and overshoot enhance. Output changes of operating variable K and a are shown 

in Figure 4b: there is evident variation at time points of t=0.03day or t=0.07day (in all aforementioned cases, 

line color in the figure corresponds to that in figure 4a), and a conclusion can be drawn in accordance with that 

in Figure 4a. 

In Figures 4 and 5, it is certain that performance of the controller is closely related to its parameters, such as 

sampling time, predictive step length, input weight value, etc. Therefore in actual use of the MPC controller, a 

trial-and-error method is usually adopted to repeatedly debug parameters of the controller and lastly determine 

its parameter configuration for an optimal balance under basic performances including response time and 

overshoot, which can help control process be more stable and reliable. 

3. Results and discussion 

Figure 6 shows control performance simulation comparison of the strategy in this paper (red solid line) and the 

PI control strategy (green dashed line). As can be seen from the figure, the strategy in this paper is superior to 

the PI control strategy in control accuracy, deviation, and response time. 

7 8 9 10 11 12 13 14
1.8

1.9

2.0

2.1

2.2

D
O

 i
n
 m

g
/L

Time in days         
7 8 9 10 11 12 13 14

1.6

1.8

2.0

2.2
D

O
 i

n
 m

g
/L

Time in days

 MPC

 PI

 

Figure 5:  Performance of the controller                         Figure 6:  Comparison of control simulation 

4. Conclusion 

This paper has put forward a dissolved oxygen control method based on MPC. The method utilized the 

aforementioned simulation software package for generating large amounts of data, acquired the state space 

model of DO value process through identification, and designed the MPC controller whose parameters are 

gradually determined by the common trial-and-error method. In simulation, it firstly conducted contrast 

verification of controller performance under different controller parameters settings so as to determine better 

controller parameters and lay the foundation for DO value process control. Secondly, it conducted control 

performance research on activated sludge process by the parameters-defined MPC controller and made a 

comparison between built-in PI control strategies based on IWA and COST Benchmark. Results show that this 

controller performs better, and the DO value becomes more stable and less undulant. The proposed method 

has two advantages: first, smaller DO activities may make the activated sludge process more stable and 

reliable and thus lead to better processing effects; second, poorer DO fluctuations would exert little load for 

heating blowing machines, which is conducive to its energy-saving operation and consequently will provide 

conditions for low-cost operation of the whole activated sludge process. 

Acknowledgments  

Research on separating characteristics and the application of water supply of ash materials from biomass 

power plant. 

Reference  

Castiglioni S., Bagnati R., Fanelli R., Pomati F., Calamari D., Zuccato E., 2006, Removal of pharmaceuticals 

in sewage treatment plants in italy, Environmental Science & Technology, 40(1), 357-63, DOI: 

10.1021/es050991m 

Lagana A., Bacaloni A., De L.I., Faberi A., Fago G., Marino A., 2004, Analytical methodologies for determining 

the occurrence of endocrine disrupting chemicals in sewage treatment plants and natural waters, Analytica 

1235



Chimica Acta, 501(1), 79–88, DOI: 10.1016/j.aca.2003.09.020 

Metcalfe C.D., Koenig B.G., Bennie D.T., Servos M., Ternes T.A., Hirsch R., 2003, Occurrence of neutral and 

acidic drugs in the effluents of canadian sewage treatment plants, Environmental Toxicology & Chemistry, 

22(12), 2872–2880, DOI: 10.1897/02-469 

Metcalfe C.D., Miao X.S., Koenig B.G., Struger J., 2003, Distribution of acidic and neutral drugs in surface 

waters near sewage treatment plants in the lower great lakes, Canada, Environmental Toxicology & 

Chemistry, 22(12), 2881-9, DOI: 10.1897/02-627 

Perrone D., Amelio M., 2016, Numerical simulation of MILD (moderate or intense low-oxygen dilution) 

combustion of coal in a furnace with different coal gun positions, International Journal of Heat and 

Technology, 34(S2), S242-S248, DOI: 10.18280/ijht.34Sp0208 

Postigo C., Alda M.J.L.D., Barceló D., 2010, Drugs of abuse and their metabolites in the ebro river basin: 

occurrence in sewage and surface water, sewage treatment plants removal efficiency, and collective drug 

usage estimation, Environment International, 36(1), 75-84, DOI: 10.1016/j.envint.2009.10.004 

Solé M., Alda M.J.L., Castillo M., Porte C., Ladegaard-Pedersen K., Barceló D., 2000, Estrogenicity 
determination in sewage treatment plants and surface waters from the catalonian area (ne spain), 
Environmental Science & Technology, 34(24), 5076-5083, DOI: 10.1021/es991335n 

Stasinakis A.S., Gatidou G., Mamais D., Thomaidis N.S., Lekkas T.D., 2008, Occurrence and fate of 

endocrine disrupters in greek sewage treatment plants, Water Research, 42(6-7), 1796-1804, DOI: 

10.1016/j.watres.2007.11.003 

Svenson A., Allard A.S., Ek M., 2003, Removal of estrogenicity in swedish municipal sewage treatment plants, 

Water Research, 37(18), 4433-43, DOI: 10.1016/S0043-1354(03)00395-6 

Tauxe-Wuersch A., Alencastro L.F.D., Grandjean D., Tarradellas J., 2005, Occurrence of several acidic drugs 

in sewage treatment plants in switzerland and risk assessment, Water Research, 39(9), 1761-72, DOI: 

10.1016/j.watres.2005.03.003 

Ternes T.A., 1998, Occurrence of drugs in German sewage treatment plants and rivers 1, Water Research, 

32(11), 3245-3260, DOI: 10.1016/S0043-1354(98)00099-2 

Ternes T.A., Kreckel P., Mueller J., 1999, Behaviour and occurrence of estrogens in municipal sewage 

treatment plants-II. Aerobic batch experiments with activated sludge, Science of the Total Environment, 

225(1-2), 91-99, DOI: 10.1016/S0048-9697(98)00335-0 

Ternes T.A., Stumpf M., Mueller J., Haberer K., Wilken R.D., Servos M., 1999, Behavior and occurrence of 

estrogens in municipal sewage treatment plants-I. Investigations in Germany, Canada and Brazil, Science 

of the Total Environment, 225(1-2), 81-90, DOI: 10.1016/S0048-9697(99)00057-1 

Vieno N., Tuhkanen T., Kronberg L., 2007, Elimination of pharmaceuticals in sewage treatment plants in 

Finland, Water Research, 41(5), 1001-1012, DOI: 10.1016/j.watres.2006.12.017 

Wagner M., Loy A., 2002, Bacterial community composition and function in sewage treatment systems, 

Current Opinion in Biotechnology,13(3), 218-27, DOI: 10.1016/S0958-1669(02)00315-4 

Yu J., Hu J., Tanaka S., Fujii S., 2009, Perfluorooctane sulfonate (pfos) and perfluorooctanoic acid (pfoa) in 

sewage treatment plants, Water Research, 43(9), 2399-2408, DOI: 10.1016/j.watres.2009.03.009 

Zorita S., Mårtensson L., Mathiasson L., 2009, Occurrence and removal of pharmaceuticals in a municipal 

sewage treatment system in the south of sweden, Science of the Total Environment, 407(8), 2760-2770, 

DOI: 10.1016/j.scitotenv.2008.12.030 

1236