Detecting the optimal condition of Zeolite utilization in industrial wastewater 1 treatment by using the fuzzy simulation model 2

12 To define optimal condition of Zeolite utilization was used the maximum industrial 13 pollution load from the industrial wastewater discharged to treatment plant lied on the industrial 14 town of Shiraz, Fars, Iran. In this study, chemical oxygen demand (COD) load and electrical 15 conductivity (EC) load from this wastewater treatment plant (WWTP) along with the parameters 16 of received industrial effluents, e.g., temperature, total suspended solid (TSS), total dissolved 17 solid (TDS) and PH were monitored. Using the several mathematical models that defined on 18 their relationships was resulted the correlation coefficients of 83% and 90% for COD with TDS, 19 and COD with TSS, respectively, thus the best regression coefficient was 0.5 under linear and 20 nonlinear forms. Autoregressive integrated moving average model (ARIMA) was also used for 21 obtaining the better results, such that the exceeded values of TSS and TDS at before day were 22 defined as input variables. In this log-time of bioreactor process, curve fitting approach and clustering analysis with and without normalized data had not improved the regression coefficient of linear and nonlinear functions. The simulation model based on fuzzy inference system (FIS) 25 has been good corresponding with the distribution of estimated and observed data of COD, so 26 that comparing such distribution of COD with line of 1:1 resulting the regression coefficient is 27 0.764. This study indicated that fixed value of the soluble solids concentration on industrial 28 wastewater discharged into treatment plant has the important role in effectiveness of Zeolite 29 filtration, i.e., it's threshold is occurred in 1746 ppm in this case study.


45
The water shortage is a growing problem in the world due to the freshwater resources 46 have been becoming not enough to satisfy the demand. Although the freshwater shortage seems 47 that the most cases of a climate-bound regional problems, it has reported in whole the world, e.g. 48 , North Africa, the Middle East, southern Europe, Australia, and the southern states of the USA. 49 Therefore, during the past few decades, there has been a growing interest in water alternative 50 sources development such as the used urban water and desalinized brackish water and seawater 51 [3]. According to UNESCO-WWAP [33] more than 70% of the water that is discharge all over 52 the world used for the agricultural irrigation, however there is a big potential for the application 53 of treated wastewater in irrigation [19]. The most important concerns in the agricultural use of countries throughout the world as reported by [26,12]. 60 The key for safe irrigation is the water quality. Many standards have been to set by 61 different institutions to control the quality of the irrigated water. The concentration and 62 composition of the dissolved constituents in water combined with the amount of water used 63 determines its quality for irrigation. Soils also vary in their capacity to resist adverse changes due 64 to the components of the water. A comprehensive water analysis will indicate its suitability for parameters, e.g., PH, T, EC, and TDS associated with COD have been measuring every day.  analyses were applied, due to the non-significant correlation between these variables.

124
The statistical techniques for analyzing the time-series data have been included the range from simple to very complex. However, the first step in such analyses is always to identify the 126 characteristics of data, thus the multiple range tests are performed by using the correlation 127 analyses to test all the pairwise comparisons among daily series means of each monitored 128 parameter. Time series analysis was applied to establish the general trend of each effluent 129 parameter. In the time series analysis, it is assumed that the data consists of a systematic pattern 130 (identifiable component) and random noise (error), which makes the pattern difficult to assess.  This study thus designed that the exceeded variables of COD or EC were the output of 141 models and other loaded factors were the inputs parameters. Since, data set of the inputs and 142 outputs factors had not the significant correlation with together, here was assumed that having 143 the significant correlation at log-time, i.e., one to three days. This idea was generated due to the 144 retention time of wastewater on zeolite channel is the key index to be an effect on the exceeded 145 water quality. Hence, the models of time series analysis, i.e., ARIMA were used, then the curve 146 fitting approach were utilized to achieve the simulation model based on the relationship between 147 the normalized data of exceeded daily COD and loaded TDS on previous day. In such process 148 needs to normalize data as following: (X max -X i ) / (X max -X min ) (1) 150 Where X max and X min are the maximum and minimum values of input or output variables.

151
The several linear and nonlinear models based on the curve fitting approach were used to find a

Results and discussion
The popular statistical analysis has usually used to study the statistical characteristic of 207 data on the wastewater variables included central tendency, variability, and shape. There have 208 the standardized skewness and standardized kurtosis that to be applied to determine whether the 209 sample having the normal distribution, the statistics outside of the range of -2 to +2 indicates 210 significant departures from normality and would tend to invalidate many of the statistical 211 procedures that normally applied on the data. The standardized skewness values outside expected 212 range for the followed variables were shown In   were more than the beyond Zeolite filtration capacity.
The correlation coefficients of COD with TDS, and COD with TSS were 83% and 90%, 242 respectively (Table 3). Although, the relationship between COD and TDS is not the significant    The five models of ARIMA compared for selecting the best time-series (Table 4), 271 resulting the one day log has been better matching on other models, so that predicted daily

284
Using the curve fitting approach with and without the normalized data to achieve any 285 function of the COD E (t) based on the TDS L (t-1) and ECE (t) with TSS L (t-1) has not improved 286 the regression coefficient for linear and nonlinear functions. The weak relationships between 287 COD E (t) and TDS L (t-1) has explained how the data distribution shown in the Fig. 4. The 288 classification analysis of data distribution based on the data normalized average K-means has 289 used in this study (Table 5).  Table 5 297 The results of clustering analysis for the loaded values of TDS and TSS in the day of t-1, and the Based on the results of clustering analysis, the ranges of CODE (t) and TDSL (t-1) 301 variables classified to the five major classes as shown in Table 5. The COD is being in low class 302 until the TDS has in moderate or high class, resulting it assumed that this trend has the fuzzy 303 behavior, and the fuzzy inference systems (FIS) can be appropriated to simulate this distribution. 304 Therefore, in the first, the membership functions of input e.g., TDS t-1 , and output e.g., COD t variables in Mamdini's approach was defined based on the average of each class. In this way, the 306 five classes e.g. very low, low, normal, high and very high have described for both variables, 307 such that the functions as "very very low" and "very very high" used to define the extreme states 308 as 0 and / or 1, respectively; each input and output variables includes the seven bell membership 309 functions (Fig. 5). The Mamdani's modeling process is generally to be needed a supervisor that 310 known as the trend training. In this study, the major classes of data distribution had supervisor 311 role, and based on their relationships, the seven fuzzy rules were defined such that the Fuzzy 312 function results of the variables of input e.g., TDS t-1 and output COD t can be established as 313 follows:

314
Rules no. 1: If TDS t-1 is very very low then COD t is very very high.
(6) 315 Rules no. 2: If TDS t-1 is very low then COD t is normal.

316
Rules no. 3: If TDS t-1 is low then COD t is high.

317
Rules no. 4: If TDS t-1 is normal then COD t is very high.

318
Rules no. 5: If TDS t-1 is high then COD t is very high.

319
Rules no. 6: If TDS t-1 is very high then COD t is low.

320
Rules no. 7: If TDS t-1 is very very high then COD t is very very low.

322
In the next step, an artificial test dataset of TDS t-1 from 0 to 1 created and used in above The results of validation process of FIS model presented in Fig. 5, has obtained by 334 comparing the measured data of COD (t) with observed data, in which it is shown in Fig. 6. This  From the Fig. 4 and Fig. 5, if the TDS L was "low" then the COD E was "high", and 343 conversely, if the TDS L was "high" then the COD E was "low", so that there had closed to the  should be controlled together.

355
The highest regression coefficient of the tested linear and non-linear models to simulate 356 the variations of EC E with of TTS L were less than 0.5, and after that used the curve fitting  Upgrading the efficiency in wastewater treatment management has the goal of decision 370 maker to model the relationship between input and output variables, due to here needs that these 371 models to evaluate treatment process. Therefore, the purpose of this study has been achieving an for the variables of input, i.e., TDS t-1 and output, i.e., COD t .

378
The Fuzzy rules could been described the data distributions of input and output variables, variables, TDS t-1 along with output, COD t has formed similar to "S" shape curve, and could been 384 thus used as a reference for the filtration capacity of Zeolite channel.

385
The certain variations range of TDS have been existing that the filtration by Zeolite could 386 been affected, it is way that beginning of this impact is defined as such critical point, i.e., the 387 threshold amount of TDS. In this point, Zeolite responsibility on the removal COD in wastewater 388 treatment is to be sufficient, for example in this study, TDS threshold was occurred in 1746 ppm