The biological treatment process (BTP) is responsible for removing chemical oxygen demand (COD) and ammonia using microorganisms present in wastewater. The BTP consumes large quantities of energy due to the transfer of oxygen using air pumps/blowers. Energy consumption in the BTP is due to low solubility of oxygen, which results in low aeration efficiency (AE). The aim of the study was to develop an AE model that can be used to monitor the performance of the BTP. Multilayer perceptron artificial neural network (MLP ANN) algorithm was used to model AE in the BTP. The performance of the AE model was evaluated using R2, mean square error (MSE), and root mean square error (RMSE). Sensitivity analysis was performed on the AE model to determine variables that drive AE. The results of the study showed that MLP ANN algorithm was able to model AE. R2, MSE, and RMSE results were 0.939, 0.0025, and 0.05, respectively, during testing phase. Sensitivity analysis results showed that temperature (34.6%), COD (21%), airflow rate (19.1%), and OTR/KLa (15.7%) drive AE. At high temperatures, the viscosity of wastewater is low which enables oxygen to penetrate the wastewater, resulting in high AE. The AE model can be used to predict the performance of the BTP, which will assist in minimizing energy consumption.

  • Aeration efficiency increases with an increase in temperature.

  • Aeration efficiency decreases with an increase in airflow rate.

  • Temperature and airflow rate are the biggest driver of energy consumption.

  • Temperature, COD, and airflow rate control aeration efficiency in the biological treatment process.

  • High temperatures and airflow rates improve volumetric mass transfer coefficient.

Graphical Abstract

Graphical Abstract
Graphical Abstract
View largeDownload slide
Graphical Abstract
Graphical Abstract
View largeDownload slide
aeration efficiency, airflow rate, COD concentration, oxygen uptake rate, temperature, volumetric mass transfer coefficient

Wastewater treatment plants (WWTPs) are used all over the world including South Africa (SA). WWTPs are responsible for purifying domestic and industrial wastewater to an acceptable effluent quality before being discharged into the environment (Żyłka et al. 2021). The biological treatment process is responsible for decomposing organic matter and oxidising inorganic matter using microorganisms present in wastewater (Kanaujiya et al. 2019). The biological treatment process consumes large quantities of energy due to the transfer of oxygen using air pumps/blowers for respiration and survival of microorganisms (Wang et al. 2016; Christoforidou et al. 2020; Siatou et al. 2020). For example, Wang et al. (2016) reported that the biological energy consumption in Germany, United States of America (USA), and SA were 0.296 kWh/m³, 0.37 kWh/m³, and 0.3–0.61 kWh/m³, respectively. The energy consumption reported in SA, USA, and Germany accounts for 0.55%, 0.6%, and 0.7%, of the annual electricity consumption in their respective countries. In Spain and Saudi Arabia, biological treatment processes consume 0.8 kWh/m³ and 1.6 kWh/m³, respectively (Christoforidou et al. 2020). This energy consumption is high and requires attention in order to achieve low energy consumption.

The high energy consumption in the biological treatment process is due to the fact that oxygen is partially soluble in wastewater, which leads to low aeration efficiency (AE) (Lewis & Whitman 1924). This means that the biological treatment process will not be supplied with sufficient oxygen for respiration of microorganisms due to the resistance of wastewater in transferring oxygen. The low solubility of oxygen in the biological treatment process is caused by the gas liquid film that prevents a gas from dissolving in a liquid as defined by the two film theory (Lewis & Whitman 1924). The viscosity of the wastewater provides the resistance to the passage of oxygen molecules in the biological treatment process, which forces plant managers/operators to aerate 24 h nonstop in order to maintain a high oxygen transfer rate/volumetric mass transfer coefficient (OTR/KLa) (Rosso & Stenstrom 2006; Garcia-Ochoa & Gomez 2009; Suresh et al. 2009). AE is defined as the ratio of OTR/KLa and the power consumed during the aeration process. AE can be defined by Equation (1).
formula
(1)

In order to achieve a high AE, high OTR/KLa must be maintained so that oxygen is supplied in the biological treatment process efficiently. To achieve high OTR/KLa, research scholars have resorted to high airflow rates in order to penetrate the thickness of the wastewater caused by impurities that provide the resistance of oxygen molecules to dissolve in wastewater (Vogelaar et al. 2000; Limpaiboon 2013; Shukla & Goel 2018; Zhou et al. 2018; Warunyuwong & Imai 2020). Utilizing high airflow rates will increase the air velocity distribution which increases the shear force in the biological treatment process. An increase in the shear force will breakdown the thickness of the wastewater making it possible for oxygen molecules to penetrate and dissolve (Ren et al. 2018). According to Zhou et al. (2018), an airflow rate of 0.5 vvm (volume of air per volume of liquid per minute) produced 13.21 kg O2/h OTR/KLa, while an airflow rate of 2 vvm produced 22.43 kg O2/h OTR/KLa. Similarly, Vogelaar et al. (2000) reported that an airflow rate of 0.15 vvm produced 22 kg O2/h OTR/KLa, while an airflow rate of 0.56 vvm produced 61 kg O2/h OTR/KLa. It is clear from the literature that high airflow rates will result in high OTR/KLa. The challenge with increasing airflow rate to obtain high OTR/KLa is that power consumption also increases, resulting in a low AE (Baylar & Bagatur 2000; Pittoors et al. 2014; Al Ba'ba'a & Amano 2017; Collivignarelli et al. 2019; Du et al. 2020; Kan et al. 2020). According to Collivignarelli et al. (2019), an airflow rate of 100 m³/h produced 0.98 kg O2/kWh AE, while an airflow rate of 150 m³/h produced 0.85 kg O2/kWh AE. This is a decrease of 13.3% in AE due to an increase of 33.3% in airflow rate. Similarly, Al Ba'ba'a & Amano (2017) reported that an airflow rate of 0.125 L/min produced 2.8 kg O2/kWh AE, while an airflow rate of 0.05 L/min produced 5.2 kg O2/kWh AE. This was a decrease of 46.2% in AE due to an increase of 60% in airflow rate. Although high airflow rates improve OTR/KLa, they do not benefit AE in the biological treatment process because when high airflow rates are applied, high power consumption are utilized by the air pump/blower. This will result in high energy consumption.

High temperatures, on the other hand, have been reported to improve OTR/KLa, which results in an improvement of AE (Krahe et al. 1996; Vogelaar et al. 2000; Chalasani & Sun 2007; Yang & Park 2012; Pappenreiter et al. 2019). This is because at high temperatures, the viscosity of the wastewater decreases, which reduces the thickness of the wastewater and allows oxygen molecules to penetrate and dissolve more rapidly, resulting in high AE (Lewis & Whitman 1924). According to Vogelaar et al. (2000), a temperature of 20 °C produced OTR/KLa of 17.4 kg O2 h − 1, and a temperature of 55 °C produced OTR/KLa of 37.4 kg O2 h − 1. This is an increase of 53.5% in OTR/KLa due to an increase of 63.6% in temperature. Similarly, Yang & Park (2012) reported that a temperature of 13 °C produced OTR/KLa of 7 kg O2/h while a temperature of 23.5 °C produced OTR/KLa of 7.8 kg O2/h. This was an increase of 10.3% in OTR/KLa due to an increase of 44.7% in temperature. Therrien et al. (2019) reported that a temperature of 10 °C produced OTR/KLa of 0.0324 kg O2/h, and a temperature of 30 °C produced OTR/KLa of 0.0376 kg O2/h. This was a 13.8% gain in OTR/KLa due to a 66.7% increase in temperature. The increase of OTR/KLa due to an increase in temperature will result in an increase of AE because no power consumption was increased. An increase in AE will lead to a reducing in energy consumption in the biological treatment process, because low airflow rates can be utilized with high temperatures to achieve a high OTR/KLa delivery. The challenge is that the normal operating temperature of wastewater is usually in the range between 18 °C and 22.5 °C (Golzar et al. 2020; Tejaswini et al. 2020). This low temperature of wastewater will result in a low OTR/KLa, which will result in low AE due to high viscosity of the wastewater. Therefore, it is essential to develop a model that can be used to monitor AE in order to achieve low energy consumption in the biological treatment process. The aim of this study was to develop an AE model that can be used to monitor the performance of the biological treatment process. The model will incorporate the effects of temperature variations and airflow rate variations since they drive the efficiency of the biological treatment process. This will assist plant managers/operators in achieving high AE while utilizing low airflow rates that reduce energy consumption in the biological treatment process.

Laboratory experiments

Laboratory experiments were carried out at the Tshwane University of Technology, Faculty of Engineering and the Built Environment, Department of Civil Engineering to determine the optimum operating temperature and airflow rate that improves AE. The airflow rates used were 5 L/min, 10 L/min, 15 L/min, 20 L/min, 25 L/min, and 30 L/min. The airflow rate range was chosen based on the fact that low airflow rate produces low OTR/KLa compared with high airflow rates. The low OTR/KLa will result in a low AE, whereas high OTR/KLa will result in high AE in the biological treatment process. The temperatures used were 15 °C, 17.5 °C, 20 °C, 22.5 °C, 25 °C, 27.5 °C, 30 °C, 32.5, and 35 °C. The temperature range was chosen based on the fact that low temperature result in high viscosity compared with high temperature. The low temperature will result in low AE, whereas high temperature will increase the AE in the biological treatment process. The biological treatment process comprised of the following components: aeration tank, dissolved oxygen (DO) meter and probe, thermostat, air pump, airflow meter, air stone disc bubble diffuser, and a digital wattmeter. The components and the schematic diagram of the biological treatment process are detailed in Table 1 and Figure 1, respectively.
Table 1

Components of the aeration unit

ComponentsTypeRemarks
Aeration tank Circular aeration tank (acrylic material) 10–15 mm thick acrylic material. Dimensions 570 × 300 mm. Total volume is 40 L. Working volume is 30 L 
Dissolved oxygen meter and probe Hanna HI98196 multi-parameter water proof meter Measures DO, pH, and ORP 
Thermostat Local manufacturer Controls temperature in the range 0 °C–40 °C 
Air pump Waterfall Resun LP100 Can supply air between 0 and 140 L/min 
Airflow meter MF5712 200 L/min digital gas air nitrogen oxygen mass flow meter Can measure airflow rate between 0 and 200 L/min 
Air stone disc bubble diffuser Growneer micro pore bubble diffuser Diffuser is 20 cm in diameter 
Digital wattmeter Geewiz (kilowatt) Measures power between 0 and 3,600 watts 
ComponentsTypeRemarks
Aeration tank Circular aeration tank (acrylic material) 10–15 mm thick acrylic material. Dimensions 570 × 300 mm. Total volume is 40 L. Working volume is 30 L 
Dissolved oxygen meter and probe Hanna HI98196 multi-parameter water proof meter Measures DO, pH, and ORP 
Thermostat Local manufacturer Controls temperature in the range 0 °C–40 °C 
Air pump Waterfall Resun LP100 Can supply air between 0 and 140 L/min 
Airflow meter MF5712 200 L/min digital gas air nitrogen oxygen mass flow meter Can measure airflow rate between 0 and 200 L/min 
Air stone disc bubble diffuser Growneer micro pore bubble diffuser Diffuser is 20 cm in diameter 
Digital wattmeter Geewiz (kilowatt) Measures power between 0 and 3,600 watts 
View Large
Figure 1
Laboratory aeration unit.
View largeDownload slide

Laboratory aeration unit.

Figure 1
Laboratory aeration unit.
View largeDownload slide

Laboratory aeration unit.

Close modal

Wastewater collection

Wastewater samples were collected from the Daspoort WWTP bioreactor number two each day. Two samples were collected, one in the morning and one in the afternoon. The collection method (morning and afternoon) was implemented to prevent any biological activity from taking place during the storage process. The APHA (2012) method was used for the collection of wastewater from the Daspoort WWTP. Non sterile powder free nitrile gloves were used to handle all equipment. A 25 L jerry can was used to collect the raw wastewater. An additional 2 L of activated sludge was collected from the return activated sludge. Wastewater samples were collected and aerated within 2 h after collection. After collection, the wastewater was stored in ice bucket at a temperature of 4 °C to make sure that no biological activities take place during transportation to the Tshwane University of Technology, Faculty of Engineering and the Built Environment, Department of Civil Engineering.

Analysis

Chemical oxygen demand (COD) was measured and analysed. The method used for analysing COD was the Standard Methods for the Examination of Water and Wastewater (APHA 2012). COD was analysed using closed reflux colorimetric (APHA 2012 method 5220). Spectrophotometer (DR3900), digestion vessels, block heater, microburet, and ampule sealer were the equipment used for the analysis.

Aeration process

Aeration process was conducted twice a day (two samples collected). Two wastewater samples were aerated per day and each sample was aerated for 4 h (hydraulic retention time at Daspoort WWTP). This means that two different temperatures (35 °C and 32.5 °C) were applied during the aeration process at one of the airflow rates (5 L/min) per day. The same process was repeated the following day applying two different temperatures (30 °C and 27.5 °C) and the same airflow rate (5 L/min). The process was then repeated on all temperatures at different airflow rates. A thermostat (Table 1) was used to increase (heat up) the temperature of wastewater from field temperature (18–22.5 °C) to the desired operating temperature. Ice cubes were used to decrease (reduce) the temperature of wastewater from field temperature to the desired operating temperature. Three samples were collected and analysed during each aeration process:

  • Initial concentration immediately after collection or before aeration process.

  • 2 h concentration.

  • 4 h concentration.

Disposal

After the completion of the aeration process, wastewater was disposed by flushing at the laboratory toilet. The disposal method allows the wastewater to go back to the WWTP.

Determination of oxygen uptake rate, oxygen transfer rate/volumetric mass transfer coefficient, and aeration efficiency

The dynamic method as described by Garcia-Ochoa & Gomez (2009) was used to determine the oxygen uptake rate (OUR) and OTR/KLa.

Oxygen uptake rate

Five steps were followed to determine OUR (Garcia-Ochoa & Gomez 2009):

  • i.

    Step one – the airflow into the aeration unit was turned off for 10 min, which was enough to prevent microorganisms from being starved of oxygen.

  • ii.

    Step two – the decrease in DO concentration was monitored during the non-aeration period using a DO meter and probe.

  • iii.

    Step three – the decrease in DO was plotted against time using Microsoft Excel.

  • iv.

    Step four – a linear regression line was used to determine the rate of change in DO concentration with time. The rate of change in DO concentration was regarded as the OUR.

  • v.
    Step five – Equation (2) defines the rate of change in DO concentration with time.
    formula
    (2)

Oxygen transfer rate/volumetric mass transfer coefficient

Seven steps were followed to determine OTR/KLa (Garcia-Ochoa & Gomez 2009).

  • i.

    Step one – turn on the air supply in the aeration unit.

  • ii.

    Step two – record the DO concentration at different time intervals.

  • iii.

    Step three – measure the average DO concentration saturation in the biological aeration unit.

  • iv.
    Step four – determine the natural logarithm of the difference between DO concentration at different time intervals and average DO concentration obtained at infinite time using Equation (3).
    formula
    (3)
  • v.
    Step five – determine the difference between time intervals using Equation (4).
    formula
    (4)
  • vi.

    Step six – produce a graph plot of Equations (3) and (4).

  • vii.

    Step seven – produce a linear regression line. The slope of the linear regression line is equal to the OTR/KLa.

Aeration efficiency

Three steps were followed to determine AE (Garcia-Ochoa & Gomez 2009).

  • i.

    Step one – determine OTR/KLa in the aeration unit. OTR/KLa will be determined as described in section 2.6.2.

  • ii.

    Step two – determine the power used during the aeration process. Power input was determined as described in Table 1, using the digital wattmeter.

  • iii.

    Step three – determine the AE. AE was determined by the quotient of OTR/KLa and power input as described in Equation (1).

Application of artificial neural network

A regression model was developed using multilayer perceptron (MLP) artificial neural network (ANN) algorithm on MATLAB programming software. Continuous data collected from laboratory experiments was used to develop the AE model. The MLP ANN architecture is shown in Figure 2.
Figure 2
Aeration efficiency MLP ANN architecture.
View largeDownload slide

Aeration efficiency MLP ANN architecture.

Figure 2
Aeration efficiency MLP ANN architecture.
View largeDownload slide

Aeration efficiency MLP ANN architecture.

Close modal
The general form of the MLP ANN algorithm is shown in Equation (5). Where X is the nodal value for the previous layer. Y is the nodal value in the current layer. f is the activation function. is the weighting factor, and b is the bias of the node.
formula
(5)
Nine steps were followed to model AE using MLP ANN algorithm.

  • The first step was to select input variables from the data collected during laboratory experiments.

  • The second step was to split the data into 70% training, 15% validation, and 15% testing.

  • The third step was to transform data into input variables using the normalization technique defined by Equation (6).
    formula
    (6)
    where is the scaled sample data point, x is the sample data point, is the minimum value in the training dataset, is the maximum value on the training dataset.

  • The fourth step was to select the MLP ANN architecture with one hidden layer and four neurons.

  • The fifth step was to initialize the weights and bias between values of zero and one.

  • The sixth step was to use the sigmoid activation function as a transfer function. The sigmoid function is defined by Equation (7) and Figure 3.
    formula
    (7)

  • The seventh step was to use the supervised gradient descent backpropagation learning algorithm to correct the weights and bias.

  • The eighth step was to evaluate the performance of the AE model using MSE, R2, and RMSE defined by Equations (8)–(10).
    formula
    (8)
    formula
    (9)
    formula
    (10)
    where is the mean squared error, is the predicted data, is the observed data, N is the number of data points, is the coefficient of determination, is the sum of squared error, and is the total sum of squares.

  • The ninth step was to perform sensitivity analysis on the AE model. A one factor at a time method was used. Each decision variable was individually adjusted by ± 10%, ± 20%, and ± 50%, while keeping the others unchanged. In addition, decision variables were checked for their direct, inverse, major, or minor changes on AE.

Figure 3
Sigmoid Function.
View largeDownload slide

Sigmoid Function.

Figure 3
Sigmoid Function.
View largeDownload slide

Sigmoid Function.

Close modal

Laboratory results

Table 2 presents Pearson correlation coefficient analysis results using a two tailed test at 95% significance while Figure 4(a)–(e) presents the relationship between AE and input parameters (COD, airflow rate, temperature, OUR, and OTR/KLa). COD (−0.043**) produced a negative correlation relation to AE as shown in Table 2. This was because high concentrations of organic matter causes bubble coalescence in the biological treatment process. Bubble coalescence results in large bubbles that are faster, reducing contact time between air bubbles and wastewater. The less contact time between air bubbles and wastewater results in low AE, since the oxygen molecules carried by the air bubbles will not be able to penetrate wastewater thickness caused by high organic matter (Fan et al. 2014; Durán et al. 2016; Odize et al. 2017). Odize et al. (2017) reported that AE decreased by 0.47% with an increase in colloidal/particulate COD from 5 mg/L to 20 mg/L. Fan et al. (2014) reported a decrease of 1.744% in AE when the organic content increased from 2.840 g/L to 10.498 g/L in the biological activated sludge process. This can also be justified by the relationship between COD and OTR/KLa that produced a negative correlation of - 0.579**, which indicates that when COD increases, OTR/KLa decreases. The relationship between AE and COD shown in Figure 4(a) shows high AE at low COD concentration.
Table 2

Pearson correlation coefficient (2 tailed) between aeration efficiency and the parameters of interest

Correlation
TemperatureOUROTR/KLaCODAirflow rateAE
Temperature Pearson Correlation 1      
OUR Pearson Correlation 0.271** 1     
OTR/KLa Pearson Correlation 0.305** 0.747** 1    
COD Pearson Correlation 0.123** −0.695** −0.579** 1   
Airflow rate Pearson Correlation 0.000 0.097** 0.229** 0.008 1  
AE Pearson Correlation 0.575** 0.206** 0.542** −0.043** −0.061** 1 
Correlation
TemperatureOUROTR/KLaCODAirflow rateAE
Temperature Pearson Correlation 1      
OUR Pearson Correlation 0.271** 1     
OTR/KLa Pearson Correlation 0.305** 0.747** 1    
COD Pearson Correlation 0.123** −0.695** −0.579** 1   
Airflow rate Pearson Correlation 0.000 0.097** 0.229** 0.008 1  
AE Pearson Correlation 0.575** 0.206** 0.542** −0.043** −0.061** 1 

**Correlation is significant at the 0.01 level (2-tailed).

*Correlation is significant at the 0.05 level (2-tailed).

View Large
Figure 4
Relationship between aeration efficiency and COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e).
View largeDownload slide

Relationship between aeration efficiency and COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e).

Figure 4
Relationship between aeration efficiency and COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e).
View largeDownload slide

Relationship between aeration efficiency and COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e).

Close modal

Airflow rate (−0.061**) produced a negative correlation relation to AE. At high airflow rates, air bubbles tend to move faster, resulting in a shorter retention time of air bubbles that contain oxygen in the biological treatment process. The shorter retention time means that less oxygen molecules will be absorbed in the wastewater, yet high power is being consumed, resulting in low AE (Ashley et al. 2008; Al Ba'ba'a et al. 2014; Collivignarelli et al. 2019). According to Ashley et al. (2008), an airflow rate of 10 L/min produced 270 kg O2/kWh AE, while an airflow rate of 40 L/min produced 180 kg O2/kWh AE. Similarly, Al Ba'ba'a et al. (2014) reported that an airflow rate of 14.4 L/min produced 64.05 kg O2/kWh AE while an airflow rate of 84 L/min produced 2.94 kg O2/kWh AE. Figure 4(b) shows a similar relationship between AE and airflow rate. This shows that high airflow rates are not beneficial in the biological treatment process. Temperature (0.575**) produced the highest positive correlation relation to AE. This was because temperature controls the viscosity of wastewater, and at high temperatures the viscosity of wastewater is low, which increases the transfer of oxygen molecules in wastewater. When oxygen transfer is high in wastewater due to low viscosity, AE will increase because less power from air pumps/blowers is required to achieve high oxygen solubility (Therrien et al. 2019). This can also be justified by the relationship between temperature and OTR/KLa, which produced a positive correlation of 0.305**. This relationship shows that when the viscosity of wastewater is low due to high temperatures, oxygen transfer increases due to a thin liquid film that allows for high oxygen transfer. Therrien et al. (2019) reported that a temperature of 10 °C produced 0.197 kg O2/kWh AE and a temperature of 30 °C produced 0.222 kg O2/kWh AE. This resulted in a 11.3% increase due to an increase in temperature of wastewater. The relationship between AE and temperature can be observed in Figure 4(c).

OUR (0.206**) produced the third highest positive correlation relation to AE. In the biological treatment process, OTR/KLa is equal OUR when airflow rate is maintained at a constant because the rate of change in dissolved oxygen concentration remains the same. In addition, high OUR indicates high oxygen consumption by microorganisms, which suggests that oxygen was available in abundance due to high OTR/KLa, resulting in high AE. This can be justified by the relationship between OTR/KLa and OUR, which produced a positive correlation of 0.747** in the biological treatment process. The graphical relationship between AE and OUR can be seen in Figure 4(d). OTR/KLa (0.542**) produced the second highest positive correlation relation to AE. When OTR/KLa increases in the biological treatment process, it indicates that more mass of oxygen is being transferred into the wastewater using less power consumption, hence AE increases. In addition, OTR/KLa is directly proportional to AE as defined by Equation (1), hence they produce a positive relationship. The relationship between AE and OTR/KLa can be observed in Figure 4(e).

Modelling performance results

MLP ANN algorithm was successful in modelling AE in the biological treatment process. Figure 5 shows the 226 observed AE data that were used as input data during the training phase and the predicted AE data from the MLP ANN AE model during the training phase. The predicted data were close to the observed data, therefore MLP ANN algorithm was able to predict the AE data during the training phase. Overall performance of the model was acceptable since it was able to predict the maximum, middle, and minimum data points accurately.
Figure 5
Observed vs predicted values during the training phase.
View largeDownload slide

Observed vs predicted values during the training phase.

Figure 5
Observed vs predicted values during the training phase.
View largeDownload slide

Observed vs predicted values during the training phase.

Close modal
Figures 6 and 7 show 49 observed data that were determined during the experiments and the predicted data from the MLP ANN algorithm during the validation and testing phase of the AE model, respectively. During the validation phase, the model predicted the minimum, middle, and maximum data accurately similar to the training phase. This means that the model generalised the data accurately as shown in Figure 6. Similarly, the model was able to predict the observed data during the testing phase. This means that the model was able to capture the observed data accurately as shown in Figure 7.
Figure 6
Observed vs predicted values during the validation phase.
View largeDownload slide

Observed vs predicted values during the validation phase.

Figure 6
Observed vs predicted values during the validation phase.
View largeDownload slide

Observed vs predicted values during the validation phase.

Close modal
Figure 7
Observed vs predicted values during the testing phase.
View largeDownload slide

Observed vs predicted values during the testing phase.

Figure 7
Observed vs predicted values during the testing phase.
View largeDownload slide

Observed vs predicted values during the testing phase.

Close modal
The prediction accuracy of the MLP ANN AE model was evaluated using R2, RMSE, and MSE shown in Table 3. The optimum MSE of 0.00075 was determined during the validation phase of the AE model. The highest MSE of 0.0025 was determined during the testing phase of the model. Although the testing phase produced the lowest performance, the model performance was still acceptable. This means that the distance between the data points and the regression line are close to each other. The difference between the highest and lowest MSE was 70%, which was a significant difference. The relationship between MSE and number of iterations is shown in Figure 8. Similar results were obtained by the following scholars (Kundu et al. 2013; Tumer & Edebali 2015; MG et al. 2018; Bekkari & Zeddouri 2019; Saleh & Kayi 2021). Saleh & Kayi (2021) applied ANN algorithm to model the prediction of COD and the model produced MSE of 0.000441. MG et al. (2018) applied ANN algorithm in modelling WWTP performance and the model produced MSE value of 0.311. Similarly, Kundu et al. (2013) applied ANN algorithm to model the biological removal of organic carbon and nitrogen in wastewater and the model produced MSE of 2.81. Bekkari & Zeddouri (2019) applied ANN algorithm to predict effluent COD in wastewater and the model produced MSE of 0.056. Tumer & Edebali (2015) applied ANN algorithm to model the biological treatment process and the model produced MSE of 0.0041.
Table 3

MLP ANN energy consumption model performance

TrainingValidationTesting
R2 0.963 0.963 0.939 
MSE 0.0012 0.00075 0.0025 
RMSE 0.0353 0.0274 0.05 
TrainingValidationTesting
R2 0.963 0.963 0.939 
MSE 0.0012 0.00075 0.0025 
RMSE 0.0353 0.0274 0.05 
View Large
Figure 8
Mean squared error performance at each iteration.
View largeDownload slide

Mean squared error performance at each iteration.

Figure 8
Mean squared error performance at each iteration.
View largeDownload slide

Mean squared error performance at each iteration.

Close modal

The blue, green, and red lines in Figure 8 indicate training, validation, and testing phases, respectively. The best line in Figure 8 indicates the lowest MSE and number of iterations during the validation phase of the model. Because the MSE was closer to zero during training, validation, and testing phases, it shows that the difference between the predicted and observed data was small. The results obtained in this study showed that MLP ANN algorithm captured the AE data accurately in the biological treatment process.

The RMSE during training, validation, and testing phase results were 0.0353, 0.0274, and 0.05, respectively, as shown in Table 3. The optimum RMSE was determined during the validation phase, which justifies the optimum MSE obtained during the validation phase. The difference between the highest and lowest RMSE was 29.4%, which was a significant difference. Overall, MLP ANN AE model performed well in all learning phases. Results obtained in this study were superior compared with the results obtained by other scholars (Hamada et al. 2018; Newhart et al. 2020; Aalami et al. 2021). Newhart et al. (2020) applied ANN algorithm to model acid disinfection performance in a WWTP and the model produced RMSE of 6.95. Similarly, Aalami et al. (2021) applied ANN algorithm to model the performance of a WWTP and the model produced RMSE of 3.93. Hamada et al. (2018) applied ANN algorithm to model the performance of a WWTP and the model produced RMSE of 29.69.

R2 also showed good performance during training, validation, and testing phases. The training phase of the model produced R2 value of 0.962 as shown in Figure 9(a). This means that 96.2% of the data points fit the model accurately. Outliers are visible in Figure 9(a), which means that some of the data points did not fit the model accurately during the training phase. Validation phase produced R2 value of 0.923 as shown in Figure 9(b). This means that 92.3% of the data points fit the model accurately. Similar to the training phase, outliers were visible which means that some of the data points did not fit the model accurately. Testing phase produced R2 value of 0.939 as shown in Figure 9(c). This was a slight drop from the validation R2. Overall, the AE model performance was acceptable.
Figure 9
R2 during training phase (a), validation phase (b), and testing phase (c).
View largeDownload slide

R2 during training phase (a), validation phase (b), and testing phase (c).

Figure 9
R2 during training phase (a), validation phase (b), and testing phase (c).
View largeDownload slide

R2 during training phase (a), validation phase (b), and testing phase (c).

Close modal

Similar results were obtained by the following scholars (Kundu et al. 2013; Tumer & Edebali 2015; Hamada et al. 2018; Bekkari & Zeddouri 2019; Ofman & Struk-Sokołowska 2019; Oulebsir et al. 2020; Alsulaili & Refaie 2021). Alsulaili & Refaie (2021) applied ANN algorithm to model five day biological oxygen demand in wastewater treatment and the model produced R2 value of 0.754. Bekkari & Zeddouri (2019) applied ANN algorithm to predict effluent COD in wastewater and the model produced R2 of 88%. Ofman & Struk-Sokołowska (2019) applied ANN algorithm in modelling nitrogen forms in wastewater and the model produced R2 of 0.99. Oulebsir et al. (2020) modelled the activated sludge energy consumption. R2 values obtained were 94.41% and 82.42% during training and testing phases. Similarly, Tumer & Edebali (2015) applied ANN algorithm to model the biological treatment process and the model produced R2 of 96%. Kundu et al. (2013) applied ANN algorithm to model the biological removal of organic carbon and nitrogen in wastewater and the model produced R2 value of 96%. Hamada et al. (2018) applied ANN algorithm to model the performance of a WWTP and the model produced R2 of 78%.

Sensitivity analysis

Figure 10 presents the results of the AE model sensitivity analysis. The biggest driver of AE was temperature (34.6%). This was expected since temperature controls the viscosity of wastewater. COD (21%) was the second highest contributor to AE. This was due to the fact that organic content creates impurities in wastewater, which will create a resistance of oxygen molecules to dissolve in the wastewater. Airflow rate (19.1%) and OTR/KLa (15.7%) contributed almost equally towards AE in the biological treatment process. The least contributor to AE model was OUR (9.6%). This implies that OUR can be eliminated as an input to predict AE. Temperature, COD, airflow rate, and OTR/KLa are more suitable as input variables in the AE model.
Figure 10
Aeration efficiency model sensitivity analysis.
View largeDownload slide

Aeration efficiency model sensitivity analysis.

Figure 10
Aeration efficiency model sensitivity analysis.
View largeDownload slide

Aeration efficiency model sensitivity analysis.

Close modal
The relationships between each MLP ANN AE output results and input variables are presented in Figure 11. The input variables were controlled at 74 mg/L, 25 L/min, 35 °C, 0.2567 mg/L, and 0.1899/h for COD, airflow rate, temperature, OUR, and OTR/KLa, respectively. The control parameters produced an AE of 0.569 kg O2/kWh. The relationship between AE and COD concentration showed an increase of up to 88.8 mg/L, beyond that the AE decreased. This was because high COD concentrations decrease the ability of oxygen molecules to dissolve in the wastewater due to high impurities, resulting in low AE. Similarly, high airflow rates beyond 25 L/min affected AE, because high airflow rate consumes high power yet producing a moderate to high OTR/KLa, resulting in low AE in the biological treatment process. This shows that high airflow rates are not beneficial to improve AE. At high temperatures (30 °C and above) AE decreased. This suggests that the optimum operating temperature that allows for high AE was 30 °C. AE increased with an increase in OUR. This indicates that when OUR increases due to an increase in microorganisms, AE increases in the biological treatment process. AE increased with an increase in OTR/KLa up to 0.1899/h. Beyond OTR/KLa of 1,899/h, AE decreased.
Figure 11
Effect of COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e) on aeration efficiency.
View largeDownload slide

Effect of COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e) on aeration efficiency.

Figure 11
Effect of COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e) on aeration efficiency.
View largeDownload slide

Effect of COD (a), airflow rate (b), temperature (c), OUR (d), and OTR (e) on aeration efficiency.

Close modal

The aim of the study was to develop an AE model that can be used to monitor the performance of the biological treatment process. The following data collected from the laboratory was used: COD, temperature, airflow rate, OUR, and OTR/KLa. Low concentration of COD showed high AE. High AE was measured at low airflow rate of 15 L/min. High temperatures (35 °C) produced the highest AE of 0.621 kg O2/kWh. The highest AE (0.621 kg O2/kWh) was measured at moderate OUR of 0.2567 mg/l. AE increased with an increase in OTR/KLa in the biological treatment process. MLP ANN algorithm was used to model the AE data. MLP ANN algorithm was successful in modelling the AE data. R2 values were 0.963, 0.963, and 0.939 for training, validation, and testing phases respectively. MSE values were 0.0012, 0.00075, and 0.0025 for the training, validation, and testing phases, respectively. RMSE values were 0.0353, 0.0274, and 0.05 for the training, validation, and testing phases, respectively. Sensitivity analysis showed that temperature (34.6%) and COD (21%) were the biggest drivers of AE in the biological treatment process. The AE model can be utilized by plant managers/operaters to improve AE which ultimately improves energy consumption in the biological treatment process.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.

Aalami
M. T.
,
Hejabi
N.
,
Nourani
V.
 &
Saghebian
S.
2021
Investigation of artificial intelligence approaches capability in predicting the wastewater treatment plant performance (Case study: Tabriz wastewater treatment plant)
.
Amirkabir Journal of Civil Engineering
53
(
3
),
15
–
15
.
doi:10.22060/CEEJ.2019.16757.6334
.
Google Scholar
 
Al Ba'ba'a
H. B.
 &
Amano
R. S.
2017
A study of optimum aeration efficiency of a lab-scale air-diffused system
.
Water and Environment Journal
31
(
3
),
432
–
439
.
https://doi.org/10.1111/wej.12261
.
Google Scholar
Crossref
Search ADS
 
Al Ba'ba'a
H. B.
,
Prada
M. A.
,
Olson
C. D.
,
Alkhalidi
A. A.
,
Amano
R. S.
 &
Li
J.
2014
An experimental study of reducing back pressure of fine air diffuser used in wastewater plants
. In:
ASME 2014 4th Joint US-European Fluids Engineering Division Summer Meeting Collocated with the ASME 2014 12th International Conference on Nanochannels, Microchannels, and Minichannels
.
American Society of Mechanical Engineers
, pp.
V01CT16A006
–
V01CT16A006
.
https://doi.org/10.1115/FEDSM2014-21203
.
Google Scholar
Crossref
Search ADS
 
American Public Health Association (APHA), American Water Works Association, Water Environment Federation
2012
Standard Methods for the Examination of Water and Waste Water
, 22nd edn.
American Public Health Association, Washington, DC, USA
.
Ashley
K. I.
,
Mavinic
D. S.
 &
Hall
K. J.
2008
Effect of orifice diameter, depth of air injection, and air flow rate on oxygen transfer in a pilot-scale, full lift, hypolimnetic aerator
.
Canadian Journal of Civil Engineering
36
(
1
),
137
–
147
.
https://doi.org/10.1139/S08-047
.
Google Scholar
Crossref
Search ADS
 
Baylar
A.
 &
Bagatur
T.
2000
Aeration performance of weirs
.
Water SA
26
(
4
),
521
–
526
.
Available from: https://hdl.handle.net/10520/AJA03784738_1609.
Google Scholar
 
Bekkari
N.
 &
Zeddouri
A.
2019
Using artificial neural network for predicting and controlling the effluent chemical oxygen demand in wastewater treatment plant
.
Management of Environmental Quality: An International Journal
.
https://doi.org/10.1108/MEQ-04-2018-0084
.
Google Scholar
 
Chalasani
G.
 &
Sun
W.
2007
Measurement of Temperature Effects on Oxygen Uptake Rate in Activated Sludge Treatment
.
Report, Michigan State University College of Engineering
,
East Lansing, MI
,
USA
, pp.
1
–
28
.
Available from
:
http://citeseerx.ist.psu.edu/viewdoc/download?doi = 10.1.1.529.6379&rep = rep1&type = pdf
.
Christoforidou
P.
,
Bariamis
G.
,
Iosifidou
M.
,
Nikolaidou
E.
 &
Samaras
P.
2020
Energy benchmarking and optimization of wastewater treatment plants in Greece
. In
Environmental Sciences Proceedings
. (Vol.
2
, No.
1
).
Multidisciplinary Digital Publishing Institute
, p.
36
.
https://doi.org/10.3390/environsciproc2020002036
.
Google Scholar
Crossref
Search ADS
 
Collivignarelli
M. C.
,
Abbà
A.
 &
Bertanza
G.
2019
Oxygen transfer improvement in MBBR process
.
Environmental Science and Pollution Research
,
1
–
11
.
https://doi.org/10.1007/s11356-019-04535-1
.
Google Scholar
 
Du
Y.
,
Chen
F.
,
Zhou
L.
,
Qiu
T.
 &
Sun
J.
2020
Effects of different layouts of fine-pore aeration tubes on sewage collection and aeration in rectangular water tanks
.
Aquacultural Engineering
89
,
102060
.
https://doi.org/10.1016/j.aquaeng.2020.102060
.
Google Scholar
Crossref
Search ADS
 
Durán
C.
,
Fayolle
Y.
,
Pechaud
Y.
,
Cockx
A.
 &
Gillot
S.
2016
Impact of suspended solids on the activated sludge non-newtonian behaviour and on oxygen transfer in a bubble column
.
Chemical Engineering Science
141
,
154
–
165
.
https://doi.org/10.1016/j.ces.2015.10.016
.
Google Scholar
Crossref
Search ADS
 
Fan
H.
,
Qi
L.
,
Liu
G. H.
,
Zhang
Y.
,
Chen
X.
 &
Wang
H.
2014
Promotion and inhibition of oxygen transfer under fine bubble aeration by activated sludge
.
Water and Environment Journal
28
(
3
),
434
–
441
.
https://doi.org/10.1111/wej.12061
.
Google Scholar
Crossref
Search ADS
 
Garcia-Ochoa
F.
 &
Gomez
E.
2009
Bioreactor scale-up and oxygen transfer rate in microbial processes: an overview
.
Biotechnology Advances
27
(
2
),
153
–
176
.
https://doi.org/10.1016/j.biotechadv.2008.10.006
.
Google Scholar
Crossref
Search ADS
PubMed
 
Golzar
F.
,
Nilsson
D.
 &
Martin
V.
2020
Forecasting wastewater temperature based on artificial neural network (ANN) technique and monte carlo sensitivity analysis
.
Sustainability
12
(
16
),
6386
.
https://doi.org/10.3390/su12166386
.
Google Scholar
Crossref
Search ADS
 
Hamada
M.
,
Adel Zaqoot
H.
 &
Abu Jreiban
A.
2018
Application of artificial neural networks for the prediction of Gaza wastewater treatment plant performance-Gaza strip
.
Journal of Applied Research in Water and Wastewater
5
(
1
),
399
–
406
.
https://doi.org/10.22126/arww.2018.874
.
Google Scholar
 
Kan
S. E.
,
Fahmi
M. R.
,
Shum
K. J.
,
Ibrahim
A. H.
,
Abidin
C. Z. A.
 &
Kusuma
B. S.
2020
Improvement of oxygen transfer efficiency in the activated sludge process
. In
IOP Conference Series: Earth and Environmental Science
. (Vol.
476
, No.
1
).
IOP Publishing
, Bristol, UK, p.
012098
.
Available from: https://iopscience.iop.org/article/10.1088/1755-1315/476/1/012098/meta.
Google Scholar
Crossref
Search ADS
 
Kanaujiya
D. K.
,
Paul
T.
,
Sinharoy
A.
 &
Pakshirajan
K.
2019
Biological treatment processes for the removal of organic micropollutants from wastewater: a review
.
Current Pollution Reports
5
(
3
),
112
–
128
.
https://doi.org/10.1007/s40726-019-00110-x
.
Google Scholar
Crossref
Search ADS
 
Krahe
M.
,
Antranikian
G.
 &
Märkl
H.
1996
Fermentation of extremophilic microorganisms
.
FEMS Microbiology Reviews
18
(
2–3
),
271
–
285
.
https://doi.org/10.1111/j.1574-6976.1996.tb00243.x
.
Google Scholar
 
Kundu
P.
,
Debsarkar
A.
 &
Mukherjee
S.
2013
Artificial neural network modeling for biological removal of organic carbon and nitrogen from slaughterhouse wastewater in a sequencing batch reactor
.
Advances in Artificial Neural Systems
2013
.
http://dx.doi.org/10.1155/2013/268064
.
Google Scholar
 
Limpaiboon
K.
2013
Influence of operating conditions and physical properties of liquid medium on volumetric oxygen transfer coefficient in a dual impeller bioreactor
.
Walailak Journal of Science and Technology (WJST)
10
(
6
),
625
–
634
.
Available from: https://103.58.148.28/index.php/wjst/article/view/472.
Google Scholar
 
MG
H. N.
,
Manilal
A. M.
,
Anand
P. G.
 &
Soloman
P. A.
2018
Artificial Neural Network Based Control of Electrocoagulation based Automobile Wastewater Treatment Plant
. In
2018 International Conference on Circuits and Systems in Digital Enterprise Technology (ICCSDET)
.
IEEE
, pp.
1
–
4
.
doi:10.1109/ICCSDET.2018.8821131
.
Google Scholar
 
Newhart
K. B.
,
Goldman-Torres
J. E.
,
Freedman
D. E.
,
Wisdom
K. B.
,
Hering
A. S.
 &
Cath
T. Y.
2020
Prediction of peracetic acid disinfection performance for secondary municipal wastewater treatment using artificial neural networks
.
ACS ES&T Water
1
(
2
),
328
–
338
.
https://doi.org/10.1021/acsestwater.0c00095
.
Google Scholar
Crossref
Search ADS
 
Odize
V. O.
,
Novak
J.
,
De Clippeleir
H.
,
Al-Omari
A.
,
Smeraldi
J. D.
,
Murthy
S.
 &
Rosso
D.
2017
Reverse flexing as a physical/mechanical treatment to mitigate fouling of fine bubble diffusers
.
Water Science and Technology
76
(
7
),
1595
–
1602
.
https://doi.org/10.2166/wst.2017.171
.
Google Scholar
Crossref
Search ADS
PubMed
 
Oulebsir
R.
,
Lefkir
A.
,
Safri
A.
 &
Bermad
A.
2020
Optimization of the energy consumption in activated sludge process using deep learning selective modeling
.
Biomass and Bioenergy
132
,
105420
.
https://doi.org/10.1016/j.biombioe.2019.105420
.
Google Scholar
Crossref
Search ADS
 
Pappenreiter
M.
,
Sissolak
B.
,
Sommeregger
W.
 &
Striedner
G.
2019
Oxygen uptake rate soft-sensing via dynamic kLa computation: cell volume and metabolic transition prediction in mammalian bioprocesses
.
Frontiers in Bioengineering and Biotechnology
7
,
195
.
https://doi.org/10.3389/fbioe.2019.00195
.
Google Scholar
Crossref
Search ADS
PubMed
 
Pittoors
E.
,
Guo
Y.
 &
Van Hulle
S. W.
2014
Oxygen transfer model development based on activated sludge and clean water in diffused aerated cylindrical tanks
.
Chemical Engineering Journal
243
,
51
–
59
.
https://doi.org/10.1016/j.cej.2013.12.069
.
Google Scholar
Crossref
Search ADS
 
Ren
J.
,
Cheng
W.
,
Wan
T.
,
Wang
M.
 &
Jiao
M.
2018
Effect of aeration rates on hydraulic characteristics and pollutant removal in an up-flow biological aerated filter
.
Environmental Science: Water Research & Technology
4
(
12
),
2041
–
2050
.
doi:10.1039/C8EW00231B
.
Google Scholar
 
Rosso
D.
 &
Stenstrom
M. K.
2006
Economic implications of fine-pore diffuser aging
.
Water Environment Research
78
(
8
),
810
–
815
.
https://doi.org/10.2175/106143006X101683
.
Google Scholar
Crossref
Search ADS
PubMed
 
Saleh
B. A.
 &
Kayi
H.
2021
Prediction of chemical oxygen demand from the chemical composition of wastewater by artificial neural networks
. In
Journal of Physics: Conference Series
, (Vol.
1818
, No.
1
).
IOP Publishing
, Bristol, UK, p.
012035
.
Available from: https://iopscience.iop.org/article/10.1088/1742-6596/1818/1/012035/meta.
Google Scholar
Crossref
Search ADS
 
Shukla
B. K.
 &
Goel
A.
2018
Study on oxygen transfer by solid jet aerator with multiple openings
.
Engineering Science and Technology, An International Journal
21
(
2
),
255
–
260
.
https://doi.org/10.1016/j.jestch.2018.03.007
.
Google Scholar
Crossref
Search ADS
 
Siatou
A.
,
Manali
A.
 &
Gikas
P.
2020
Energy consumption and internal distribution in activated sludge wastewater treatment plants of Greece
.
Water
12
(
4
),
1204
.
https://doi.org/10.3390/w12041204
.
Google Scholar
Crossref
Search ADS
 
Suresh
S.
,
Srivastava
V. C.
 &
Mishra
I. M.
2009
Techniques for oxygen transfer measurement in bioreactors: a review
.
Journal of Chemical Technology & Biotechnology: International Research in Process, Environmental & Clean Technology
84
(
8
),
1091
–
1103
.
https://doi.org/10.1002/jctb.2154
.
Google Scholar
Crossref
Search ADS
 
Tejaswini
E.
,
Babu
G. U. B.
 &
Rao
A. S.
2020
Effect of temperature on effluent quality in a biological wastewater treatment process
.
Chemical Product and Process Modeling
15
(
1
).
https://doi.org/10.1515/cppm-2019-0018
.
Google Scholar
 
Therrien
J. D.
,
Vanrolleghem
P. A.
 &
Dorea
C. C.
2019
Characterization of the performance of venturi-based aeration devices for use in wastewater treatment in low-resource settings
.
Water SA
45
(
2
),
251
–
258
.
Available from: https://hdl.handle.net/10520/EJC-15975b7367.
Google Scholar
 
Tumer
A. E.
 &
Edebali
S.
2015
An artificial neural network model for wastewater treatment plant of Konya
.
International Journal of Intelligent Systems and Applications in Engineering
3
(
4
),
131
–
135
.
https://doi.org/10.18201/ijisae.65358
.
Google Scholar
Crossref
Search ADS
 
Vogelaar
J. C. T.
,
Klapwijk
A.
,
Van Lier
J. B.
 &
Rulkens
W. H.
2000
Temperature effects on the oxygen transfer rate between 20 and 55 C
.
Water Research
34
(
3
),
1037
–
1041
.
https://doi.org/10.1016/S0043-1354(99)00217-1
.
Google Scholar
Crossref
Search ADS
 
Wang
H.
,
Yang
Y.
,
Keller
A. A.
,
Li
X.
,
Feng
S.
,
Dong
Y. N.
 &
Li
F.
2016
Comparative analysis of energy intensity and carbon emissions in wastewater treatment in USA, Germany, China and South Africa
.
Applied Energy
184
,
873
–
881
.
https://doi.org/10.1016/j.apenergy.2016.07.061
.
Google Scholar
Crossref
Search ADS
 
Warunyuwong
P.
 &
Imai
T.
2020
Study of air–water interface generator as oxygen transfer enhancer in diffused aeration system
.
Environmental Technology
,
1
–
7
.
https://doi.org/10.1080/09593330.2020.1830182
.
Google Scholar
 
Yang
H. C.
 &
Park
S. K.
2012
Oxygen transfer characteristics of an ejector aeration system
.
International Journal of Fluid Machinery and Systems
5
(
1
),
10
–
17
.
https://doi.org/10.5293/IJFMS.2012.5.1.010
.
Google Scholar
Crossref
Search ADS
 
Żyłka
R.
,
Karolinczak
B.
 &
Dąbrowski
W.
2021
Structure and indicators of electric energy consumption in dairy wastewater treatment plant
.
Science of the Total Environment
782
,
146599
.
https://doi.org/10.1016/j.scitotenv.2021.146599
.
Google Scholar
Crossref
Search ADS
PubMed
 
© 2022 The Authors
This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence (CC BY-NC-ND 4.0), which permits copying and redistribution for non-commercial purposes with no derivatives, provided the original work is properly cited (http://creativecommons.org/licenses/by-nc-nd/4.0/).