Modeling the zeolite channels effect on the effluent of treatment plant in Shiraz industrial town

Experimental
Wastewater station and monitoringWater Reclamation Plant is located in Shiraz industrial town, Fars, Iran. Wastewater of 1100 small and medium industrial units are collected through the sewage system and treated in this station. Its capacity is 2500 m3 per day but now, loaded wastewater is only about 1200 to 1500 m3 per day. Treatment Methods is completely biological and chemical purification method does not use in reactors. Logon and stilling pond with anaerobic system, UABR system, UASB, Selector, SBR and wetland are the parts of this plant. Biological treatment starts from anaerobic lagoons, that wastewater fixation done there, afterward gas and methane produced as a result of bacteria’s activity. At this point, pollution concentration decreased and the amount of COD reduced. In order to wastewater could achieve suitable retention time to provide sufficient time for the biological response, two lagoons and a stilling pond designed between the lagoons that is Plug Flow reactor type. This system will reduce energy consumption and operating costs. Depth of lagoons is about 5 to 6 meters and stilling pool depth is 120 cm in stilling pond sewage find enough time to complete the biological process. Wastewater inters to lagoon1initially and after crossing the stilling pool sheds to the lagon2. In stilling pool sewage find enough time to complete the biological process. Sludge deposited along the way and not transferred to the next unit, sludge are drains from lagoons once a week.
DataChanges in effluent before and after crossing the zeolite channel were measured according intended indices in the laboratory by regular sampling (90 samples during 3 month). PH, temperature, TSS, TDS, EC, and COD of wastewater that loaded and exceeded to zeolite channels were measured. Same analysis methods were applied to determine the effluent characteristics.
Modeling processIn order to model the effect of zeolite channels on the wastewater characteristics at first the relationship between daily values of (EC) and COD of exceeding effluent from Shiraz industrial town treatment plant after crossing the zeolite channel according other examined factors of loaded wastewater was studied. For this purpose, values of indicators such as PH, temperature, TSS, TDS, of loading wastewater was used as known values and factors such as EC, and COD were considered as unknown values. Correlation coefficient was used to predict wastewater quality parameters. Then curve fitting approach was used to consider several linear and non- linear models to simulate this correlation. This method was used to find the correlation coefficients between exceeding COD and EC and other loaded factors from industrial wastewater. The results of forecasting models in the calibration phase and the results of their validation process were used to select the appropriate model.
Performance of models, were studied according to Root Mean Square Error (RMSE). Then, curve fitting approach used to achieve a simulation model on relationship between the normalized data of exceeded daily COD and loaded TDS on previous day. Several linear and non linear models based on curve fitting approach used to find a simulation model for defining the relationship between the normalized data of exceeded daily COD and loaded TDS at previous day. Then cluster analysis and the fuzzy inference system applied to improve the simulation of this relationship. The best correlation coefficient and p-vale were obtained between the exceeded values of EC and the loaded values of TSS for each day. Equations could be shown as follow:CODE (t)= F(TDSL(t))(2)
ECE(t)= F(TSSL(t)) (3)
The correlation coefficients of COD with TDS, and COD with TSS were 83% and 90%, respectively. Because measuring TDS is easier than the TSS then equation 2 was selected to estimate COD.
ARIMA models results offered the log time of one day have the best relationship than other models. Daily changes in TSS and TDS according to their value of the one day before predicted. This model predicted EC(E) and COD(t) through TSS(L) and TDS(t-1). As a result, the decision maker could be having the optimization analysis by means of practical strategies to change the collection network or Zeolite filtration capacity.
CODE(t)= F(TDS L(t-1))(4) ECE(t)= F (TSS L(t-1)) (5)
After that curve fitting approach with and without normalized data were used, to achieve a function of CODE(t) based on TDSL(t-1) and ECE(t) with TSSL(t-1). These approaches couldn’t improve the regression coefficient of each linear and nonlinear function too. Weak relationships between CODE(t) and TDSL(t-1) is the cause of how data distribution which are illustrated in figure3. As shown in table 4 it was assumed that this trend has the fuzzy behavior and using the fuzzy inference systems (FIS) model was appropriate to simulate this distribution. Therefore, at first the membership functions of input (TDS (t-1)) and output (COD (t)) variables in Mamdini approach was defined based on the average of each class.
Result and discussionWith the study purposes of achieving the appropriate model of variability of TDS(L), TSS(L), and exceeded COD(E), EC(E), no significant correlation coefficient were obtained by analyzing approaches such as multivariate, curve fitting with linear and nonlinear models, time series, and clustering with and without normalized data. Therefore, the results of classification were utilized to define the mean and range of each membership function for input variables TDS(t-1) and the output COD(t).
As noticed, fuzzy roles describe data distributions of input and output variables which other deterministic models couldn’t consider subject matter. Feature of fuzzy roles in this point caused a good conformity with estimated and measured values of exceeded COD(t) with R2 = 0.76. As a result, fuzzy behavior assumption of data distribution was admitted. The result of simulation model showed that data distribution of input variables TDS(t-1) and the output COD(t) function was similar to S shape curve. It could be refer to Zeolite channels filtering capacity.
FIS simulation model had good corresponding with distribution of measured and observed data of COD (t) R2 = 0.76. As seen in figurer 3 and 4, Zeolite filtering capacity may be affected by other factors. Low input of TDS observed when rainfall occurred, so entering wastewater content and loaded TDS affected by runoff.
The result of simulation model showed that a certain range of TDS exists in which Zeolite could be effective. In this study, TDS threshold occurred in 1746ppm in which COD is equal to 52. Result obtained in this study indicated that, fixed value of soluble solids concentration on mass transfer of wastewater entering the treatment plant had important role in the effectiveness of Zeolite filtration.
As seen in figurer 4c and 5, while loaded TDS was low and high, then exceeded COD was high and low or closed to standard level, respectively. This result indicated that Zeolite filtering capacity maybe affected by other factors. Low input of TDS observed when rainfall occurred, so entering wastewater content and loaded TDS affected by runoff. In the other word, while the volume of wastewater increased during this period, TDS values decreased.
In this condition, loaded wastewater was more than daily filtering capacity of Zeolite Channel. Although TDS values were low, the volume of wastewater entering to the treatment plant was high then Zeolite channel could not operate its ion exchange capacity appropriately. Its filtering operation was better when a few wastewaters with more TDS value entered to the Zeolite channel. Finally, this result could be achieved that to manage wastewater treatment loaded amount of pollutants and mass transfer of wastewater trough Zeolite channel should be controlled together.
The best regression coefficient of EC(E) with the TSS(L) was less than 0.5. Curve fitting approaches could not improve the regression coefficient of each linear and nonlinear function too. Based on the results of clustering analysis the ranges of ECE(t) and TSSL(t-1) variables classified into three major classes. FIS simulation model had not good corresponding with distribution of measured and observed data of ECE (t).