Ozone dosage in most water treatment plants is operated by determining the ozone concentration with the experience of the operation. In this case, it is not economical. This study selected the factors affecting residual ozone concentration and attempted to estimate the optimum amount of hydrogen peroxide dosage for the control of the residual ozone concentration by developing a model for the prediction of the residual ozone concentration. The prediction formulas developed in this study can quickly respond to the environment of water quality and surrounding environmental factors, which change in real time, so it is judged that they could be used for the operation of the optimum ozone process, and the control of ozone dosage could be used as a new method in controlling the concentration of ozone dosage and the concentration of residual ozone.

## INTRODUCTION

Recently, the emission of new toxic substances caused by the advancement and diversification of industrial structure and rapid population growth have led to pollution of water sources, and taste and odor-causing compounds have been generated in water sources by the growth of algal bloom as a natural phenomenon. In order to remove it, ozone and activated carbon processes as an advanced water treatment technology are introduced to previous sedimentation and filtration processes.

As ozone forecasting is available around us on the day when we are exposed to strong ultraviolet radiation since the Ozone Forecasting System was recently introduced in Korea, a small amount of ozone is around our daily lives. Although ozone is not an unusual chemical substance and beneficial when an appropriate amount of ozone exists in the air or liquid, ozone shows toxicity when it exists in large quantities. Ozone toxicity is caused by strong oxidizing power and plays a role in generating free radicals or peroxides and upsetting the balance of bioactive substances ( Kang *et al.* 2002).

Therefore, the injection of an appropriate amount of ozone makes it possible to secure the original disinfection capacity of ozone and at the same time to realize economic operations. In addition, the reduction of discharge ozone into air makes it possible to reduce health risk of operators and to prevent working environment from worsening. For reference, ozone concentration is set at less than 0.06 ppm of 8-hour average and less than 0.1 ppm of 1-hour average in environmental standards according to the current Framework Act on Environmental Policy.

As mentioned above, the adverse effects of the ozone process affect not only the problems caused by discharge ozone due to the deaeration of dissolved ozone but also the reduction of the following process, activated carbon's adsorption capacity. Thus, there is a need to inject an appropriate minimum amount of ozone.

Most purification plants are operated by deciding injected ozone concentrations based on the feedback of the operational experience according to concentrations of taste and odor-causing compounds in raw water flowing into purification plants or dissolved ozone concentrations at the rear of the ozone contact basin.

In such cases, excessive or insufficient ozone can be injected because it is not possible to expect the injection of an appropriate amount of ozone. It is highly probable that they are uneconomical and cannot exert the original disinfection capacity in an increase in discharge ozone and deficient ozone injection.

Therefore, this study aims to analyze the factors affecting the ozone injection rate through correlation analysis based on physical, chemical, and biological factors such as raw water and weather conditions affecting the ozone process and at the same time to realize the preemptive operation without depending on previous operation methods or feedback methods by developing a predictive model of ozone injection rate to determine the optimal injected ozone concentration.

Moreover, this study calculated the appropriate amount of hydrogen peroxide (H_{2}O_{2}) to control residual ozone concentration by selecting influencing factors affecting the residual ozone concentration and developing a predictive model of ozone injection rate.

## MATERIAL AND METHODS

Purification Plant Y managed and operated by City S was selected as a research area of this study. Purification Plant Y has a facility design capacity of 600,000 m^{3}/day, uses raw water from Intake Station P, and produces purified water to provide it to over 1.7 million citizens by introducing the first advanced water treatment plant in City S. The advanced water treatment of City S consists of membrane filtration, ozone contact basin, and activated carbon bed.

This study conducted a time series analysis using the data of the raw water quality of Purification Plant Y for one year in 2014 and analyzed the hourly and seasonal characteristics of the raw water quality in the study area. In addition, this study examined the occurrence and characteristics of representative taste and odor-causing compounds such as 2-MIB and geosmin by time series analysis.

In order to examine the extent to which 2-MIB and geosmin were removed by process for a year in 2014, coagulation, precipitation, and filtration processes were regarded as one process. This study measured concentrations of 2-MIB and geosmin after the filtration process. Subsequently, this study calculated the removal efficiency of the ozone process by measuring the concentrations of 2-MIB and geosmin in the treated water after the ozone process.

This study performed a correlation analysis between raw water quality and injected ozone concentration to develop a predictive model of ozone injection rate and performed a correlation analysis between weather conditions and injected ozone concentration. Based on the results of the correlation analyses, this study developed a predictive model for the optimal injected ozone concentration with and without taste and odor-causing compounds by stepwise multiple regression analysis.

As a prelude to the calculation of an appropriate injection amount of hydrogen peroxide to control the residual ozone concentration, this study developed a predictive model for injected ozone concentrations using multiple regression equations and considered the range of ozone injection rate to meet the operational criteria based on the developed predictive model for injected ozone concentrations.

Acidity . | Temperature . | ||||
---|---|---|---|---|---|

∼ 5 °C . | 5–10 °C . | 10–15 °C . | 15–20 °C . | 20 °C ∼ . | |

∼pH 7.0 | 150 | 110 | 70 | 45 | 30 |

pH 7.0–pH 7.5 | 85 | 65 | 40 | 25 | 15 |

pH 7.5 ∼ | 50 | 40 | 25 | 15 | 10 |

Acidity . | Temperature . | ||||
---|---|---|---|---|---|

∼ 5 °C . | 5–10 °C . | 10–15 °C . | 15–20 °C . | 20 °C ∼ . | |

∼pH 7.0 | 150 | 110 | 70 | 45 | 30 |

pH 7.0–pH 7.5 | 85 | 65 | 40 | 25 | 15 |

pH 7.5 ∼ | 50 | 40 | 25 | 15 | 10 |

This study calculated the injected hydrogen peroxide concentration using the operational data such as residual ozone concentrations, pH, and temperature and predicted injected hydrogen peroxide concentrations by combining the predicted residual ozone concentration and this equation.

## RESULTS AND DISCUSSION

A cross correlation analysis was conducted on daily mean values between the items of raw water quality. The results shown in Table 2 were obtained using 362 analytical data except 3 days without measurement records. As a result of the correlation analysis, water temperature was highly correlated with turbidity and weakly correlated with pH and residual chlorine. Turbidity was highly correlated with pH and weakly correlated with residual chlorine. There was a very strong correlation between pH, alkalinity, and electrical conductivity. It showed that they were highly correlated with residual chlorine. In addition, there was a quite strong correlation between alkalinity and electrical conductivity.

N = 362
. | Water temperature . | Turbidity . | pH . | Alkalinity . | Residual chlorine . | Electrical conductivity . |
---|---|---|---|---|---|---|

Water temperature | 1 | |||||

Turbidity | 0.456 | 1 | ||||

pH | 0.320 | 0.403 | 1 | |||

Alkalinity | 0.092 | 0.245 | 0.880 | 1 | ||

Residual chlorine | 0.397 | 0.372 | 0.410 | 0.209 | 1 | |

Electrical conductivity | −0.080 | 0.198 | 0.863 | 0.956 | 0.209 | 1 |

N = 362
. | Water temperature . | Turbidity . | pH . | Alkalinity . | Residual chlorine . | Electrical conductivity . |
---|---|---|---|---|---|---|

Water temperature | 1 | |||||

Turbidity | 0.456 | 1 | ||||

pH | 0.320 | 0.403 | 1 | |||

Alkalinity | 0.092 | 0.245 | 0.880 | 1 | ||

Residual chlorine | 0.397 | 0.372 | 0.410 | 0.209 | 1 | |

Electrical conductivity | −0.080 | 0.198 | 0.863 | 0.956 | 0.209 | 1 |

The characteristics of taste and odor-causing compounds were analyzed using 2-MIB and geosmin concentrations measured 192 times for a year in 2014. Geosmin concentrations rapidly increased in the summer. Overall, it showed low concentrations except it. Although 2-MIB concentrations varied widely between July and September, they showed similar tendencies in general.

This study conducted a correlation analysis between taste and odor-causing compounds, other items of water quality, and meteorological factors. As a result, 2-MIB was weakly correlated with water and air temperatures and poorly correlated with other water quality items and meteorological factors. Geosmin was weakly correlated with water temperature, alkalinity, electrical conductivity, and atmospheric pressure. Among them, alkalinity and electrical conductivity showed a negative correlation as a negative value Table 3.

N = 192
. | Water temp. . | Turbidity . | pH . | Alkalinity . | Residual Choline . | Electrical conductivity . | Temp. . | Atmosperic pressure . | Humidity . |
---|---|---|---|---|---|---|---|---|---|

2-MIB | 0.410 | 0.108 | 0.109 | −0.156 | 0.063 | −0.267 | 0.352 | −0.117 | 0.074 |

Geosmin | 0.313 | −0.038 | 0.062 | −0.350 | −0.173 | −0.344 | 0.300 | −0.327 | 0.279 |

N = 192
. | Water temp. . | Turbidity . | pH . | Alkalinity . | Residual Choline . | Electrical conductivity . | Temp. . | Atmosperic pressure . | Humidity . |
---|---|---|---|---|---|---|---|---|---|

2-MIB | 0.410 | 0.108 | 0.109 | −0.156 | 0.063 | −0.267 | 0.352 | −0.117 | 0.074 |

Geosmin | 0.313 | −0.038 | 0.062 | −0.350 | −0.173 | −0.344 | 0.300 | −0.327 | 0.279 |

This study regarded coagulation, precipitation, and filtration processes as one process and calculated the removal efficiency by measuring the concentrations of 2-MIB and geosmin in filtered water and then comparing them with the concentrations of 2-MIB and geosmin included in raw water. Moreover, this study evaluated the removal efficiency of taste and odor-causing compounds in the ozone process by measuring the concentrations of 2-MIB and geosmin included in effluent water after the ozone process and then comparing them with the concentrations of 2-MIB and geosmin included in filtered water. Conversely, this study selected only the measured data of concentrations of taste and odor-causing compounds in filtered water and effluent water in the ozone process among the 192 analytical data used in this study. The data with 0 ng/L of the measured concentration of taste and odor-causing compounds in filtered water from were excluded this analysis because it is not possible to calculate the removal efficiency of the ozone process as a subsequent process.

2-MIB showed the average removal efficiency of 8% in coagulation, precipitation, and filtration processes and removal efficiency of approximately 82% in the ozone process. It is considered that the concentration of 2-MIB increased in coagulation precipitation and filtration processes because of the coagulation of the scattered 2-MIB in these processes or other reasons. Additionally, it is considered possible to prevent concentrations from increasing by shortening the cleaning cycle by process. However, because these values were abnormal due to operational errors, they were excluded from the effective data when conducting a correlation analysis. As was the case in geosmin. On the other hand, 2-MIB showed low removal efficiency in coagulation, precipitation, and filtration processes based on the above results. Therefore, it is considered that the results will be used as data capable of confirming the need for advanced water treatment such as ozone process. This study developed a predictive model for ozone injection rate without taste and odor-causing compounds and another predictive model for ozone injection rate with taste and odor-causing compounds. Using multiple regression equations of SPSS 18.0.

In order to calculate the predictive model for usual ozone injection rate, this study measured taste and odor-causing compounds among daily mean 365 data and then excluded the days with concentrations. Subsequently, the days with 0 of ozone injection rate were additionally excluded. Then, the standard deviation was calculated. The data with more than 2.5 of absolute value of the standard deviation were regarded as an outlier and then deleted. Finally, 151 data were used to construct predictive models. A multiple regression analysis was conducted by stepwise selection after inputting these data, and the correlation coefficient of the sixth model was very high, 0.858. As a result of analysis of variance, the significant probability was 0.000, which is smaller than the significant level of 0.05. The derived regression equation was statistically significant because it denies the null hypothesis that a regression equation is a curve. It was considered that there was no risk of multi-collinearity because all the tolerance limit values were over 0.1 in the collinearity statistics and VIF value was less than 10 Tables 4 and 5.

Model 6 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 3.495 | 4 | 0.874 | 101.926 | 0.000 |

Residual | 1.252 | 146 | 0.009 | ||

Sum | 4.747 | 150 |

Model 6 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 3.495 | 4 | 0.874 | 101.926 | 0.000 |

Residual | 1.252 | 146 | 0.009 | ||

Sum | 4.747 | 150 |

Model 6 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | −0.181 | 0.095 | – | −1.899 | 0.049 | – | – | |

X_{1} | Residual Chlorine | 2.948 | 0.343 | 0.445 | 8.585 | 0.000 | 0.673 | 1.486 |

X_{2} | Water temperature | 0.024 | 0.002 | 0.897 | 11.780 | 0.000 | 0.312 | 3.209 |

X_{3} | Electrical conductivity | 0.005 | 0.001 | 0.779 | 6.068 | 0.000 | 0.110 | 9.122 |

X_{4} | Alkalinity | −0.015 | 0.003 | −0.587 | −5.518 | 0.000 | 0.160 | 6.261 |

Model 6 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | −0.181 | 0.095 | – | −1.899 | 0.049 | – | – | |

X_{1} | Residual Chlorine | 2.948 | 0.343 | 0.445 | 8.585 | 0.000 | 0.673 | 1.486 |

X_{2} | Water temperature | 0.024 | 0.002 | 0.897 | 11.780 | 0.000 | 0.312 | 3.209 |

X_{3} | Electrical conductivity | 0.005 | 0.001 | 0.779 | 6.068 | 0.000 | 0.110 | 9.122 |

X_{4} | Alkalinity | −0.015 | 0.003 | −0.587 | −5.518 | 0.000 | 0.160 | 6.261 |

Model 3 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 10.142 | 3 | 3.381 | 133.031 | 0.000 |

Residual | 4.574 | 180 | 0.025 | – | |

Sum | 14.716 | 183 | – | – |

Model 3 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 10.142 | 3 | 3.381 | 133.031 | 0.000 |

Residual | 4.574 | 180 | 0.025 | – | |

Sum | 14.716 | 183 | – | – |

Model 3 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | 19.934 | 1.729 | – | 11.528 | 0.000 | – | – | |

X_{1} | Atmospheric pressure | −0.019 | 0.002 | −0.560 | −11.370 | 0.000 | 0.712 | 1.405 |

X_{2} | Geosmin | 0.009 | 0.001 | 0.337 | 7.049 | 0.000 | 0.754 | 1.327 |

X_{3} | Residual chlorine | 1.596 | 0.451 | 0.152 | 3.540 | 0.001 | 0.935 | 1.069 |

Model 3 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | 19.934 | 1.729 | – | 11.528 | 0.000 | – | – | |

X_{1} | Atmospheric pressure | −0.019 | 0.002 | −0.560 | −11.370 | 0.000 | 0.712 | 1.405 |

X_{2} | Geosmin | 0.009 | 0.001 | 0.337 | 7.049 | 0.000 | 0.754 | 1.327 |

X_{3} | Residual chlorine | 1.596 | 0.451 | 0.152 | 3.540 | 0.001 | 0.935 | 1.069 |

In order to calculate the predictive model for usual residual ozone concentration 186 out of 193 data except the outliers were used when taste and odor-causing compounds did not occur. A multiple regression analysis was conducted by stepwise selection after inputting these data, and the correlation coefficient of the fourth model was very high, 0.858. As a result of analysis of variance, the significant probability was 0.000, which is smaller than the significant level of 0.05. The derived regression equation was statistically significant because it denies the null hypothesis that a regression equation is a curve. It was considered that there was no risk of multi-collinearity because all the tolerance limit values were over 0.1 in the collinearity statistics and VIF value was less than 10 Tables 8 and 9.

Model 4 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 0.063 | 4 | 0.016 | 0.01395 | 0.000 |

Residual | 0.022 | 180 | 0.000 | 0.01264 | |

Sum | 0.085 | 184 | – | 0.01116 |

Model 4 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 0.063 | 4 | 0.016 | 0.01395 | 0.000 |

Residual | 0.022 | 180 | 0.000 | 0.01264 | |

Sum | 0.085 | 184 | – | 0.01116 |

Model 4 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | 0.122 | 0.018 | – | 6.841 | 0.000 | – | – | |

X_{1} | Ozone dosage | 0.067 | 0.004 | 0.968 | 17.425 | 0.000 | 0.475 | 2.107 |

X_{2} | Temperature | −0.001 | 0.000 | −0.559 | −8.727 | 0.000 | 0.357 | 2.801 |

X_{3} | Residence time | −0.005 | 0.001 | −0.376 | −6.274 | 0.000 | 0.407 | 2.459 |

X_{4} | Filtered water turbidity | −0.314 | 0.093 | −0.148 | −3.368 | 0.001 | 0.753 | 1.327 |

Model 4 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | 0.122 | 0.018 | – | 6.841 | 0.000 | – | – | |

X_{1} | Ozone dosage | 0.067 | 0.004 | 0.968 | 17.425 | 0.000 | 0.475 | 2.107 |

X_{2} | Temperature | −0.001 | 0.000 | −0.559 | −8.727 | 0.000 | 0.357 | 2.801 |

X_{3} | Residence time | −0.005 | 0.001 | −0.376 | −6.274 | 0.000 | 0.407 | 2.459 |

X_{4} | Filtered water turbidity | −0.314 | 0.093 | −0.148 | −3.368 | 0.001 | 0.753 | 1.327 |

Also this study developed a prediction model of residual ozone concentration with taste and odor-causing compounds. 181 out of 193 data except the outliers were used when taste and odor-causing compounds did occur. The correlation coefficient of the fifth model was very high, 0856. As a result of analysis of variance, the significant probability was 0.000, which is smaller than the significant level of 0.05. The derived regression equation was statistically significant because it denies the null hypothesis that a regression equation is a curve Tables 10 and 11.

Model 5 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 0.050 | 5 | 0.010 | 96.898 | 0.000 |

Residual | 0.018 | 176 | 0.000 | – | |

Sum | 0.069 | 181 | – | – |

Model 5 . | Sum of squares . | Degree of freedom . | Mean square . | F . | Significance probability . |
---|---|---|---|---|---|

Regression Model | 0.050 | 5 | 0.010 | 96.898 | 0.000 |

Residual | 0.018 | 176 | 0.000 | – | |

Sum | 0.069 | 181 | – | – |

Model 5 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | −0.432 | 0.199 | – | −2.170 | 0.031 | – | – | |

X_{1} | Ozone dosage | 0.069 | 0.004 | 1.072 | 17.116 | 0.000 | 0.386 | 2.589 |

X_{2} | Atmospheric pressure | 0.001 | 0.000 | 0.219 | 2.698 | 0.008 | 0.229 | 4.370 |

X_{3} | Filtered water turbidity | −0.287 | 0.086 | −0.151 | −3.354 | 0.001 | 0.745 | 1.342 |

X_{4} | Residence time | −0.004 | 0.001 | −0.301 | −4.873 | 0.000 | 0.396 | 2.526 |

X_{5} | Temperature | −0.001 | 0.000 | −0.402 | −4.497 | 0.000 | 0.189 | 5.282 |

Model 5 . | Unstandardized coefficient . | Standardized coefficient . | t . | Significance probability . | Collinearity statistics . | |||
---|---|---|---|---|---|---|---|---|

B . | Standard error . | Beta . | Tolerance . | VIF . | ||||

Constant | −0.432 | 0.199 | – | −2.170 | 0.031 | – | – | |

X_{1} | Ozone dosage | 0.069 | 0.004 | 1.072 | 17.116 | 0.000 | 0.386 | 2.589 |

X_{2} | Atmospheric pressure | 0.001 | 0.000 | 0.219 | 2.698 | 0.008 | 0.229 | 4.370 |

X_{3} | Filtered water turbidity | −0.287 | 0.086 | −0.151 | −3.354 | 0.001 | 0.745 | 1.342 |

X_{4} | Residence time | −0.004 | 0.001 | −0.301 | −4.873 | 0.000 | 0.396 | 2.526 |

X_{5} | Temperature | −0.001 | 0.000 | −0.402 | −4.497 | 0.000 | 0.189 | 5.282 |

## CONCLUSIONS

First, this study evaluated the removal efficiency of taste and odor-causing compounds by water purification process. As a result, 2-MIB showed removal efficiency of approximately 8% in coagulation, precipitation, and filtration processes and removal efficiency of approximately 82% in the ozone process. In addition, geosmin showed removal efficiency of about 19% in coagulation, precipitation, and filtration processes and removal efficiency of about 79% in the ozone process. Based on the results, it was confirmed that advanced water treatment such as ozone process must be performed to remove taste and odor-causing compounds.

Second, this study developed two prediction equations as a regression equation applicable with and without taste and odor-causing compounds. Residual chlorine concentration in raw water, water temperature, electrical conductivity, and alkalinity were adopted to the usual predictive model as an explanatory variable. The correlation coefficient with the actual ozone injection rate was very high, 0.858. On the other hand, atmospheric pressure, geosmin concentration, and residual chlorine concentration in raw water were adopted to the predictive model with taste and odor-causing compounds. The correlation coefficient with the actual ozone injection rate was very high, 0.830. It was confirmed that they could be fully used in the actual water purification process.

Third, this study predicted an appropriate injected hydrogen peroxide concentration by predicting the residual ozone concentration of effluent water in the ozone process. First, in order to predict the residual ozone concentration, this study developed two predictive models applicable with and without taste and odor-causing compounds.

Injected ozone concentration, atmospheric temperature, residence time, and turbidity of filtered water were adopted to the usual predictive model as an explanatory variable. The correlation coefficient with the actual residual ozone concentration was very high, 0.858. Injected ozone concentration, atmospheric pressure, turbidity of filtered water, residence time, and atmospheric temperature were adopted to the prediction equation with taste and odor-causing compounds as an explanatory variable. The correlation coefficient with the actual residual ozone concentration was very high, 0.856.

Conversely, this study developed a prediction equation of injected hydrogen peroxide concentrations by combining the developed equation of injected hydrogen peroxide concentrations and prediction equation of residual ozone concentrations.

The developed prediction equations in this study can respond quickly to changing water quality environment and surrounding environmental factors. Therefore, it is considered possible to use them to operate the ozone process. Furthermore, it is expected that the regulation of ozone injection will be used as a new method in previous control methods for concentrations of injected ozone and residual ozone.

## ACKNOWLEDGEMENTS

This study was funded by the Seoul Business Agency as Seoul R&BD Program (PS150002).