Abstract

It is highly essential that municipal wastewater is treated before its discharge and reuse in order to meet the standard requirements for safe marine life and for farming and industries. It is beneficial to use reclaimed water, since availability of fresh water is inadequate. An investigation was conducted on the Jamnagar Municipal Corporation Sewage Treatment Plant (JMC-STP) to develop a feedforward artificial neural network (FF-ANN) model. It is an alternate for the modelling/ prediction of JMC-STP to circumvent over the versatile physical, chemical, and biological treatment process simulations. The models were developed to predict effluent quality parameters through influent characteristics. The parameters are pH, biochemical oxygen demand (BOD), chemical oxygen demand (COD), total suspended solids (TSS), total Kjeldahl nitrogen (TKN), ammonium nitrogen (AN) and total phosphorus (TP). The correlation coefficient RTRAINING and RALL were calculated for all parametric models. The MAD (mean absolute deviation), MSE (mean square error), RMSE (root mean square error) and MAPE (mean absolute percentage error) were evaluated for FF-ANN models. This proves to be a useful tool for the plant management to optimize the treatment quality as it enhances the performance and reliability of the plant. The simulation results were validated through the measured values.

ABBREVIATIONS

     
  • AI

    Artificial Intelligence

  •  
  • AN

    Ammonium Nitrogen

  •  
  • ANN

    Artificial Neural Network

  •  
  • BOD

    Biochemical Oxygen Demand

  •  
  • COD

    Chemical Oxygen Demand

  •  
  • FF-ANN

    Feedforward Artificial Neural Network

  •  
  • ICEAS

    Intermittent Cycle Extended Aeration System

  •  
  • JMC-STP

    Jamnagar Municipal Corporation Sewage Treatment Plant

  •  
  • LOGSIG

    Log Sigmoid Transfer Function

  •  
  • MAD

    Mean Absolute Deviation

  •  
  • MAPE

    Mean Absolute Percentage Error

  •  
  • MLD

    Million Liters per Day

  •  
  • MLP

    Multi-Layer Perceptron

  •  
  • MLSS

    Mixed Liquor Suspended Solids

  •  
  • NHN

    Number of Hidden Neurons

  •  
  • Ninp

    Number of Input Neurons

  •  
  • Nout

    Number of Output Neurons

  •  
  • R

    Correlation Coefficient

  •  
  • RMSE

    Root Mean Square Error

  •  
  • SBR

    Sequential Batch Reactor

  •  
  • TKN

    Total Kjeldahl Nitrogen

  •  
  • TP

    Total Phosphorus

  •  
  • TSS

    Total Suspended Solids

INTRODUCTION

Reclaimed water from the wastewater treatment process is identified as the major source of water for various agricultural and industrial engineering applications in regions with scarce fresh water (Verlicchi et al. 2012). The use of the treated municipal sewage is of utmost significance, particularly in India where dry land accounts for about 57% of total cultivated area and there is a shortage of fresh water for domestic use (Singh et al. 2018). If the untreated municipal wastewater is discharged into a natural water body, it leads to contamination. As a result, the fresh water bodies get polluted and the contamination level adversely affects the marine life (Gazette of India 1986). Rapid growth in urban population is one of the major reasons for the rise in fresh water demand and increase in the generation of waste water. As per the Central Pollution Control Board (CPCB) report, an estimated 61,754 million liters per day (MLD) domestic wastewater is generated in major cities of India whereas the wastewater treatment capacity is only 22,963 MLD. If we compare it with the industrial wastewater, only 37.18% of it is treated before discharge (CPCB 2016).

It is important to prevent the discharge of untreated sewage in Rangmati and Nagmati rivers as it continues to accumulate near the seashore of Jamnagar in Gulf of Kutch. The seashore attached to Jamnagar is a part of Marine National Park (MNP), therefore the discharge of untreated sewage affects the marine life. The MNP authorities initiated the establishment of the sewage treatment plant with Gujarat State Pollution Control Board (GSPCB) and implemented it through JMC, Jamnagar (Thakker 2013).

Intermittent cycle extended aeration system (ICEAS)-sequential batch reactor (SBR) is selected for the municipal wastewater treatment. It is an energy efficient wastewater treatment system with high treatment efficiency (Delgado San Martín et al. 2014). This type of plant is selected for large municipal wastewater treatment all over the world. The treatment processes of the domestic wastewater are complex and depend upon the influent quality as they need proper monitoring and operational control of the system (Loos et al. 2013). Researchers have developed different artificial neural network (ANN)-based models for prediction of the water quality (Mundi et al. 2018), but the prediction is limited to biochemical oxygen demand (BOD) and chemical oxygen demand (COD) (Civelekoglu et al. 2009; Abyaneh 2014).

Earlier, some researchers worked on the nonlinearity and uncertainty associated with the wastewater treatment processes simulation (Mingzhi et al. 2009). The ANN modeling of wastewater faces challenges such as data, temporal nature of the process and model optimization due to the extremely nonlinear and dynamic behaviour of the process. Wastewater treatment comprises physical, chemical and biological processes. Data collection is performed through instruments in a large volume. The data is often noisy, uncertain and incomplete. Hence, the collected data sets are pre-processed through suitable pre-processing algorithms. Different machine learning algorithms are used to obtain an accurate model. As models are highly nonlinear and dynamic in nature, artificial intelligence (AI)-based algorithms are used to select the best model structure. The AI models are introduced as the data compiled processes, which may not identify the complete physics and chemistry associated with the wastewater treatment processes. The data compiled processes simply work as a black box model providing the functional relationships between system inputs and outputs. The AI-based wastewater treatment system models can efficiently control the nonlinearity and intricacy of a system; as the ANN model overcomes the limitations of numerical and mathematical models developed earlier for the wastewater treatment system (Picos-Benítez et al. 2017).

In recent years, researchers have used the AI models for modeling and simulation of the wastewater treatment processes to predict the treatment quality. Earlier, it was simulated for the limited number of influent and effluent parameters, i.e. pH, BOD and COD, without considering multifaceted physical equations and mechanism reactions (Lee et al. 2002; Pai et al. 2007; Erdirencelebi & Yalpir 2011). Recent studies concentrate on prediction, optimization and designing of various treatment technologies through ANNs. The ANNs use archived data for effectually predicting measured parameters of the non-linear biological treatment process of a synthetic or real stream (Parthiban et al. 2007; Chen et al. 2008). Several previous studies have proven the efficacy of ANN models to envisage the wastewater treatment quality. Among others, a fuzzy neural network is used for modeling the effluent COD of activated sludge process (Tomida et al. 1999). ANN is used to foretell the COD, suspended solids (SS) and mixed liquor suspended solids (MLSS) for Ankara Central Wastewater Treatment Plant (Güçlü & Dursun 2010). Multilayer feed-forward network is used to forecast the process performance of the activated sludge process, in addition to the total suspended solids (TSS), COD and sludge volume index (Banaei et al. 2013).

Modeling and simulation of biological water and wastewater treatment processes are performed in the presence of various microalgae, bacteria, microbes, yeasts, anaerobic sludge, aerated submerged biofilms using ANN (Khataee & Kasiri 2011). The wastewater treatment plant performance is guessed through black-box modeling (Mjalli et al. 2007). With automated architectural processing, ANN is applied to model the activated sludge process (Moral et al. 2008). A hybrid ANN is used as a software sensor for optimal control of a wastewater treatment process (Choi & Park 2001). In this study, supervised learning feedforward ANN model was developed to predict the effluent quality without entering into the physical and chemical processes. This is a black box modeling to gauge the treated wastewater quality for reuse or discharge.

In this paper, seven prominent effluent parameters for discharge and reuse of treated wastewater are predicted, thus reducing the number of experiments, efforts, cost and time. ANN model uses the parameters of pH, BOD, COD, TSS, total Kjeldahl nitrogen (TKN), ammonium nitrogen (AN) and total phosphorus (TP) for increasing the reliability of the effluent quality prediction. The influence of the number of hidden nodes in the ANN model has been explored on the prediction results. The number of hidden nodes in the ANN model is predicted for each parameter through the trial and error method.

MATERIALS AND METHODS

The sewage treatment plant at Jamnagar

Jamnagar is one of the coastal industrial cities in the state of Gujarat in India. The Sewage treatment plant which was developed under a pilot project to protect the coastal resources/marine biodiversity under sustainable development and coastal region conservation of Indian coastal area, is operational since August 2016. It is located in Navagam area of Jamnagar, five kilometers away from the city centre. The sewage from Jamnagar city is collected at Gandhi Nagar and Navagam through gravity, after which it is pumped to Jamnagar Municipal Corporation Sewage Treatment Plant (JMC-STP). The location of the JMC-STP is showcased in Figure 1. An STP is developed with a capacity of 70 MLD and designed with a peak factor of 2.25 (JMC 2013).

Figure 1

Google map view of Jamnagar Municipal Corporation Sewage Treatment Plant.

Figure 1

Google map view of Jamnagar Municipal Corporation Sewage Treatment Plant.

The sewage treatment processes including the primary, secondary and tertiary treatment at the JMC-STP are shown in Figure 2. The secondary treatment process uses ICEAS-SBR (JMC 2013). The treatment quality, influent and effluent characteristics of the JMC-STP are regularly tested and monitored by the operation and maintenance (O & M) team working at the plant. The data is collected and reported regularly to the JMC and the GSPCB authorities.

Figure 2

Treatment scheme of JMC wastewater treatment plant.

Figure 2

Treatment scheme of JMC wastewater treatment plant.

Wastewater quality analysis

Sewage quality parameters considered for this study are pH, BOD, COD, TSS, TKN, AN, and TP as depicted in Table 1. The standard analytical methods have been used to analyze the sewage quality parameters for the study (Standards 1988). The pH has an impact on the microbial makeup of a wastewater system. Most biological wastewater treatment is accomplished at pH range of 6.5–8.5, which is where a majority of our common environmental microbes thrive. In wastewater sample, BOD to COD ratio gives level of biodegradability. If BOD to COD ratio is >0.5, it is biodegradable, ratio of BOD to COD between 0.3 and 0.5 shows slow action and biodegradability and if BOD to COD ratio is <0.3, it is non-biodegradable. It also increases the removal rate of nitrogen and phosphorus pollutants in sewage (Akcin et al. 2005).

Table 1

Variation of influent and effluent quality parameters at Jamnagar WWTP from January 2017 to December 2017

  N total Mean Standard deviation Sum Minimum Median Maximum 
Influent pH 180 7.38 0.13 88.59 7.23 7.42 7.52 
COD (mg/L) 180 193.39 39.91 2320.68 146.00 178.40 253.33 
BOD − 5day (mg/L) 180 98.20 5.69 1178.36 91.79 96.65 112.05 
TSS (mg/L) 180 222.67 54.62 2672.09 161.40 210.73 297.69 
TKN (mg/L) 180 33.17 1.17 398.08 31.64 33.00 35.09 
Ammonium nitrogen (mg/L) 180 18.55 0.39 222.59 17.86 18.67 19.15 
Total phosphorus (mg/L) 180 0.78 0.06 9.32 0.63 0.79 0.86 
Effluent pH 180 7.64 0.10 91.68 7.48 7.69 7.74 
COD (mg/L) 180 64.63 8.94 775.52 46.25 64.27 77.33 
BOD − 5day (mg/L) 180 12.70 0.99 152.39 11.19 13.01 13.84 
TSS (mg/L) 180 15.53 1.37 186.41 14.07 15.00 18.47 
TKN (mg/L) 180 4.66 0.64 55.93 3.85 4.65 5.88 
Ammonium nitrogen (mg/L) 180 3.41 0.08 40.95 3.27 3.42 3.51 
Total phosphorus (mg/L) 180 0.38 0.04 4.53 0.30 0.38 0.44 
  N total Mean Standard deviation Sum Minimum Median Maximum 
Influent pH 180 7.38 0.13 88.59 7.23 7.42 7.52 
COD (mg/L) 180 193.39 39.91 2320.68 146.00 178.40 253.33 
BOD − 5day (mg/L) 180 98.20 5.69 1178.36 91.79 96.65 112.05 
TSS (mg/L) 180 222.67 54.62 2672.09 161.40 210.73 297.69 
TKN (mg/L) 180 33.17 1.17 398.08 31.64 33.00 35.09 
Ammonium nitrogen (mg/L) 180 18.55 0.39 222.59 17.86 18.67 19.15 
Total phosphorus (mg/L) 180 0.78 0.06 9.32 0.63 0.79 0.86 
Effluent pH 180 7.64 0.10 91.68 7.48 7.69 7.74 
COD (mg/L) 180 64.63 8.94 775.52 46.25 64.27 77.33 
BOD − 5day (mg/L) 180 12.70 0.99 152.39 11.19 13.01 13.84 
TSS (mg/L) 180 15.53 1.37 186.41 14.07 15.00 18.47 
TKN (mg/L) 180 4.66 0.64 55.93 3.85 4.65 5.88 
Ammonium nitrogen (mg/L) 180 3.41 0.08 40.95 3.27 3.42 3.51 
Total phosphorus (mg/L) 180 0.38 0.04 4.53 0.30 0.38 0.44 

ANN model

The attractiveness of the ANN is to model the nonlinear data set. In ANN modelling, one hidden layer of multilayer perceptron can act as a universal function approximator. The desirability of the ANN model comes from the extraordinary information processing features to the biological processes such as learning, heftiness, culpability and failure tolerance, nonlinearity, high parallelism, ability to handle vague data, and their proficiency to generalize (Hajmeer et al. 2000).

The values of the weight and bias are first fixed by training the model. Here, the ANN model is used for the prediction. The multi-layer perceptron artificial neural network (MLP-ANN) model is the most popular type of feedforward neural network that learns from patterns (Mjalli et al. 2007; Lou & Zhao 2012). The ANN model was developed using MATLAB 2018. The ANN model architecture deployed to model JMC-STP's effluent quality is displayed in Figure 3. It is a standard feed-forward back propagation neural network which comprises of one input layer with seven input parameters, one hidden layer and one output layer with a single output parameter. The training of the ANN model is done through log-sigmoid transfer function; where the weights are adjusted on the neurons. The LOGSIG transfer function for the training of the ANN model is such that it takes the input values from −∞ to +∞ and produces the compressed response in the range of 0 to 1.

Figure 3

ANN feedforward back propagation network.

Figure 3

ANN feedforward back propagation network.

The LOGSIG transfer function is differentiable. That is the reason it is used in the multilayer perceptron models that are trained through back propagation algorithm (Dorofki et al. 2012). The activation function choosen for the output neuron is linear, as given in Figure 3. These layers are connected through the neurons having their specific weight, which also symbolises their strength. The network is trained with the help of training data set containing input vector and output vector. It is also trained to set the parameters’ weight and biases with the specified transfer function to approximate the output vector very close to the real values at the application of input vector.

Performance evaluation of ANN model

To assess prediction performance of the FF-ANN model developed for JMC-STP effluent quality correlation coefficient (R) for training set and all data set, mean absolute deviation (MAD), root mean square error (RMSE) and the mean absolute percentage error (MAPE) were calculated using Equations (1)–(4) (Djeddou & Achour 2015). These values were analyzed for the optimization to select the best model out of all the simulated models: 
formula
(1)
 
formula
(2)
 
formula
(3)
 
formula
(4)
where: : correlated value; : observed value; : number of observations; : maximum of observed value; : minimum of observed value; and : average of observed values.

RESULTS AND DISCUSSION

The JMC-STP experimental data was used to construct the FF-ANN models. For this purpose, the data set of 180 samples was obtained from the JMC-STP O & M team The data set was divided into two groups. 70% of the data points were used for training and the remaining 30% of data points were used for testing and validation of the trained FF-ANN model. The monthly mean values of measured pH, BOD, COD, TSS, TKN, AN and TP for raw and treated sewage from January 2017 to December 2017 are illustrated in Figure 4, which indirectly present the treatment efficiency of the JMC-STP.

Figure 4

Mean values of measured influent and effluent pH, BOD, COD, TSS, TKN, AN and TP from January 2017 to December 2017.

Figure 4

Mean values of measured influent and effluent pH, BOD, COD, TSS, TKN, AN and TP from January 2017 to December 2017.

The correlation among raw (influent) sewage parameters and treated (effluent) sewage quality parameters of JMC-STP was established. Table 2 exhibits the correlation matrix of influent and effluent parameters. It reveals the correlation coefficient between each effluent parameter and the influent parameters. The FF-ANN model was developed for the JMC-STP with seven input parameters which are pH, BOD, COD, TSS, TKN, AN, and TP. The output layer consisted of one neuron related to the output variables pH, BOD, COD, TSS, TKN, AN, and TP. In the function estimation, one hidden layer is selected (Basheer 2000). In the ANN design, the number of hidden layers and hidden nodes selection is a typical task. Therefore, to determine the appropriate value, hit and trial runs are performed. There is no prior knowledge available regarding the number of hidden nodes. The design of the ANN model with a higher number of hidden nodes leads to noise, created due to overparameterization. It leads to a poor generalization of the untrained data set and takes too much time to train the model (Hajmeer et al. 2000). Some general rules suggest that the number of hidden nodes in the ANN should be in between NINP and 2NINP + 1. The literature also suggests that the ANN architecture should resemble a pyramid with: 
formula
(5)
Table 2

Correlation matrix among raw (influent) sewage quality parameters and the treated (effluent) sewage quality parameters of JMC-STP

 pH-i COD-i BOD-i TSS-i TKN-i AN-i TP-i pH-e COD -e BOD-e TSS-e TKN-e AN-e TP-e 
pH-i 1.000              
COD-i −0.845 1.000             
BOD-i −0.552 0.626 1.000            
TSS-i −0.872 0.880 0.752 1.000           
TKN-i −0.690 0.873 0.806 0.821 1.000          
AN-i −0.244 0.574 0.565 0.442 0.776 1.000         
TP-i −0.597 0.634 0.512 0.709 0.618 0.397 1.000        
pH-e 0.954 −0.734 −0.386 −0.740 −0.487 −0.034 −0.438 1.000       
COD-e −0.723 0.871 0.725 0.808 0.893 0.678 0.766 −0.557 1.000      
BOD-e −0.791 0.854 0.574 0.825 0.713 0.377 0.877 −0.674 0.870 1.000     
TSS-e 0.656 −0.543 −0.330 −0.621 −0.325 −0.038 −0.538 0.607 −0.562 −0.757 1.000    
TKN-e −0.828 0.614 0.648 0.820 0.666 0.387 0.559 −0.722 0.656 0.562 −0.486 1.000   
AN-e −0.100 0.146 0.630 0.295 0.458 0.513 0.254 0.051 0.258 0.008 0.342 0.443 1.000  
TP-e −0.494 0.262 0.215 0.367 0.294 0.071 0.743 −0.438 0.378 0.501 −0.194 0.531 0.302 1.000 
 pH-i COD-i BOD-i TSS-i TKN-i AN-i TP-i pH-e COD -e BOD-e TSS-e TKN-e AN-e TP-e 
pH-i 1.000              
COD-i −0.845 1.000             
BOD-i −0.552 0.626 1.000            
TSS-i −0.872 0.880 0.752 1.000           
TKN-i −0.690 0.873 0.806 0.821 1.000          
AN-i −0.244 0.574 0.565 0.442 0.776 1.000         
TP-i −0.597 0.634 0.512 0.709 0.618 0.397 1.000        
pH-e 0.954 −0.734 −0.386 −0.740 −0.487 −0.034 −0.438 1.000       
COD-e −0.723 0.871 0.725 0.808 0.893 0.678 0.766 −0.557 1.000      
BOD-e −0.791 0.854 0.574 0.825 0.713 0.377 0.877 −0.674 0.870 1.000     
TSS-e 0.656 −0.543 −0.330 −0.621 −0.325 −0.038 −0.538 0.607 −0.562 −0.757 1.000    
TKN-e −0.828 0.614 0.648 0.820 0.666 0.387 0.559 −0.722 0.656 0.562 −0.486 1.000   
AN-e −0.100 0.146 0.630 0.295 0.458 0.513 0.254 0.051 0.258 0.008 0.342 0.443 1.000  
TP-e −0.494 0.262 0.215 0.367 0.294 0.071 0.743 −0.438 0.378 0.501 −0.194 0.531 0.302 1.000 
The Kolmogorov theorem suggests that the number of hidden nodes should always be equal to or less than NINP + 1 
formula
(6)
where: NINP: input parameters and NOUT: output parameters.

Equations (5) and (6) are used to select the range of the number of hidden neurons. The algorithm employed for determination of the number of hidden neurons for FF-ANN JMC-STP model is shown in Figure 5. The optimum model is selected for prediction of output quality parameters by changing the number of hidden nodes from 4 to 7. It is preferred to have less hidden layers in the neural network. It reduces the problem of overfitting in the data set. To avoid the overfitting and better generalization, cross-validation of the data set is performed. The ANN model training is stopped when minimum MAD value for cross-validation data set is achieved. In this manner, the optimized ANN model architecture was selected.

Figure 5

Flow chart of ANN modeling.

Figure 5

Flow chart of ANN modeling.

The training of the ANN models was achieved by using the Levenberg-Marquardt algorithm. The designing of the network model was developed by putting the weight and bias values on the neurons connecting different layers of the network, using the LOGSIG transfer function. The number of nodes in the hidden layer was calculated by using the maximum number of the coefficient of correlation using trial and error method. After calculating the range for the selection of the number of hidden neurons in the hidden layer, it was tested for the entire range to achieve the highest prediction performance. The performance statistics of the ANN models for predicting effluent quality pH, BOD, COD, TSS, TKN, AN and TP, with a different number of hidden neurons for the training, was calculated. Initially,Kolmogorov theorem was used to find the range for the number of hidden neurons. The number of hidden neurons in the ANN model was set to 4 and modeled up to 7. Table 3 presents the ANN models of performance statistics for the effluent parameters prediction.

Table 3

Performance indexes of the optimum ANN for pH, BOD, COD, TSS, TKN, AN, and TP prediction models

Parameter name ANN Optimum Model RTRAINING RALL MAD MSE RMSE MAPE 
pH ANN 7_5_1 0.836 0.816 0.075 0.009 0.093 0.977 
BOD ANN 7_6_1 0.738 0.649 1.271 2.530 1.591 10.236 
COD ANN 7_5_1 0.701 0.656 11.350 199.326 14.118 25.440 
TSS ANN 7_4_1 0.516 0.457 2.387 8.707 2.951 16.399 
TKN ANN 7_5_1 0.736 0.670 0.930 1.481 1.217 22.241 
AN ANN 7_6_1 0.504 0.493 0.245 0.090 0.301 7.256 
TP ANN 7_6_1 0.771 0.748 0.053 0.004 0.064 15.745 
Parameter name ANN Optimum Model RTRAINING RALL MAD MSE RMSE MAPE 
pH ANN 7_5_1 0.836 0.816 0.075 0.009 0.093 0.977 
BOD ANN 7_6_1 0.738 0.649 1.271 2.530 1.591 10.236 
COD ANN 7_5_1 0.701 0.656 11.350 199.326 14.118 25.440 
TSS ANN 7_4_1 0.516 0.457 2.387 8.707 2.951 16.399 
TKN ANN 7_5_1 0.736 0.670 0.930 1.481 1.217 22.241 
AN ANN 7_6_1 0.504 0.493 0.245 0.090 0.301 7.256 
TP ANN 7_6_1 0.771 0.748 0.053 0.004 0.064 15.745 

Optimum correlation for training RTRAINING is observed for pH, ANN 7_5_1 model is having highest RTRAINING = 0.836 followed by TP and BOD. The nadir RTRAINING = 0.504 is observed for AN. Similarly, highest RALL = 0.816 for pH prediction in ANN 7_5_1 model is followed by TP and TKN which is compared to the results presented by Astray et al. 2016. The minimum RALL = 0.457 is for TSS. Minimum MAD = 0.053, MSE = 0.004, and RMSE = 0.064 are observed for TP with ANN 7_6_1 model. Nadir value of MAPE = 0.977% is seen in pH with ANN 7_5_1 model which is better than the MAPE calculated in Hawari & Alnahhal 2016 and Civelekoglu et al. 2009. The ANN models for the prediction of the effluent quality parameters having the best prediction performance were thus identified. However, slight changes were observed during estimation. Using these ANN models, the nonlinear relationship between the influent and effluent quality parameters are developed (Nadiri et al. 2018). There is a strong correlation between the measured values of the influents and the predicted values of effluents as demonstrated through ANN modeling of JMC-STP. The predicted values and the observed values of the effluent parameters pH, BOD, COD, TSS, TKN, AN and TP from January to December 2017 have been depicted in Figure 6. For the ease of presentation and simplicity, only 60 samples are displayed.

Figure 6

Observed values and the predicted values of effluent using optimized ANN models for pH, BOD, COD, TSS, TKN, AN and TP.

Figure 6

Observed values and the predicted values of effluent using optimized ANN models for pH, BOD, COD, TSS, TKN, AN and TP.

CONCLUSIONS

In this study, an FF-ANN model was developed to predict the treatment quality of the JMC-STP Jamnagar, Gujarat, India. FF-ANN model uses seven influent parameters which increase the reliability of the effluent quality prediction. The performance of the JMC-STP was predicted through the analysis of the predicted effluent from the influent parameters. This research presents that the FF-ANN with training function ‘TRAINLM’, adaption learning function ‘LEARNGDM’, performance function ‘MSD’ and single hidden layer with ‘TANSIG’ transfer function models which can predict pH, BOD, COD, TSS, TKN, AN and TP parameters for the JMC-STP. The influence of the number of hidden nodes in the ANN model was explored on the basis of prediction results. The number of hidden nodes in the ANN model was optimized through the trial and error method for each parameter. The most suitable ANN model was ANN_7_5_1 for pH with RTRAINING = 0.836, RALL = 0.816, MAD = 0.075, MES = 0.009, RMSE = 0.093, and MAPE = 0.977%. However, after pH prediction, ANN_7_6_1 for BOD with RTRAINING = 0.738, RALL = 0.649, MAD = 1.271, MES = 2.530, RMSE = 1.591, and MAPE = 10.236% were observed. There is a strong correlation between the measured values of the influents and the predicted values of effluents, as demonstrated through ANN modeling

ACKNOWLEDGEMENTS

The contributions of Centre of Material Science & Energy Studies of the LNMIIT, Jaipur and the generous support for study and data collection from JMC-STP authorities and ESSAR Projects Limited India are gratefully acknowledged.

REFERENCES

REFERENCES
Akcin
G.
,
Alp
Ö.
,
Gulyas
H.
&
Büst
B.
2005
Characteristic, analytic and sampling of wastewater
.
EMWater E-Learning Course, Lesson A1, pp. 1–49
.
Astray
G.
,
Soto
B.
,
Lopez
D.
,
Iglesias
M. A.
&
Mejuto
J. C.
2016
Application of transit data analysis and artificial neural network in the prediction of discharge of Lor River, NW Spain
.
Water Science and Technology
73
,
1756
1767
.
Banaei
F. K.
,
Zinatizadeh
A. A. L.
,
Mesgar
M.
&
Salari
Z.
2013
Dynamic performance analysis and simulation of a full scale activated sludge system treating an industrial wastewater using artificial neural network
.
International Journal of Engineering, Transactions B: Applications
26
,
465
472
.
Basheer
I. A.
2000
Selection of methodology for neural network modeling of constitutive hystereses behavior of soils
.
Computer-Aided Civil and Infrastructure Engineering
15
,
445
463
.
Chen
S. H.
,
Jakeman
A. J.
&
Norton
J. P.
2008
Artificial intelligence techniques: an introduction to their use for modelling environmental systems
.
Mathematics and Computers in Simulation
78
,
379
400
.
CPCB
2016
Central Pollution Control Board
.
Ministry of Environment and Forests, Govt. of India
,
New Delhi
,
India
.
Delgado San Martín
J. A.
,
Cruz Bournazou
M. N.
,
Neubauer
P.
&
Barz
T.
2014
Mixed integer optimal control of an intermittently aerated sequencing batch reactor for wastewater treatment
.
Computers and Chemical Engineering
71
,
298
306
.
Djeddou
M.
&
Achour
B.
2015
The use of a neural network technique for the prediction of sludge volume index in municipal wastewater treatment plant
.
Larhyss Journal
24
,
351
370
.
Dorofki
M.
,
Elshafie
A. H.
,
Jaafar
O.
,
Karim
O. A.
&
Mastura
S.
2012
Comparison of artificial neural network transfer functions abilities to simulate extreme runoff data
.
2012 International Conference on Environment, Energy and Biotechnology
33
,
39
44
.
Gazette of India
1986
The Environment (protection) rules, 1986, Govt. of India. Ministry of Environment and Forests (Department of Environment, Forest and Wildlife)
.
Güçlü
D.
&
Dursun
Ş.
2010
Artificial neural network modelling of a large-scale wastewater treatment plant operation
.
Bioprocess and Biosystems Engineering
33
,
1051
1058
.
Hajmeer
M. N.
,
Basheer
I. A.
,
Marsden
J. L.
&
Fung
D. Y. C.
2000
New approach for modeling generalized microbial growth curves using artificial neural networks
.
Journal of Rapid Methods and Automation in Microbiology
8
,
265
283
.
Hawari
A. H.
&
Alnahhal
W.
2016
Predicting the performance of multi-media filters using artificial neural networks
.
Water Science and Technology
74
,
2225
2233
.
JMC, J
2013
BEP for JMC-STP. Jamnagar, India
.
Lee
D. S.
,
Jeon
C. O.
,
Park
J. M.
&
Chang
K. S.
2002
Hybrid neural network modeling of a full-scale industrial wastewater treatment process
.
Biotechnology and Bioengineering
78
,
670
682
.
Loos
R.
,
Carvalho
R.
,
António
D. C.
,
Comero
S.
,
Locoro
G.
,
Tavazzi
S.
,
Paracchini
B.
,
Ghiani
M.
,
Lettieri
T.
,
Blaha
L.
,
Jarosova
B.
,
Voorspoels
S.
,
Servaes
K.
,
Haglund
P.
,
Fick
J.
,
Lindberg
R. H.
,
Schwesig
D.
&
Gawlik
B. M.
2013
EU-wide monitoring survey on emerging polar organic contaminants in wastewater treatment plant effluents
.
Water Research
47
,
6475
6487
.
MATLAB
2018
.
The MathWorks, Inc.
Natick, MA, USA
.
Mingzhi
H.
,
Ma
Y.
,
Jinquan
W.
&
Yan
W.
2009
Simulation of a paper mill wastewater treatment using a fuzzy neural network
.
Expert Systems with Applications
36
,
5064
5070
.
Mjalli
F. S.
,
Al-Asheh
S.
&
Alfadala
H. E.
2007
Use of artificial neural network black-box modeling for the prediction of wastewater treatment plants performance
.
Journal of Environmental Management
83
,
329
338
.
Mundi
G. S.
,
Zytner
R. G.
,
Warriner
K.
&
Gharabaghi
B.
2018
Predicting fruit and vegetable processing wash-water quality
.
Water Science and Technology
2017
,
256
269
.
Nadiri
A. A.
,
Shokri
S.
,
Tsai
F. T. C.
&
Asghari Moghaddam
A.
2018
Prediction of effluent quality parameters of a wastewater treatment plant using a supervised committee fuzzy logic model
.
Journal of Cleaner Production
180
,
539
549
.
Pai
T. Y.
,
Tsai
Y. P.
,
Lo
H. M.
,
Tsai
C. H.
&
Lin
C. Y.
2007
Grey and neural network prediction of suspended solids and chemical oxygen demand in hospital wastewater treatment plant effluent
.
Computers and Chemical Engineering
31
,
1272
1281
.
Picos-Benítez
A. R.
,
López-Hincapié
J. D.
,
Chávez-Ramírez
A. U.
&
Rodríguez-García
A.
2017
Artificial intelligence based model for optimization of COD removal efficiency of an up-flow anaerobic sludge blanket reactor in the saline wastewater treatment
.
Water Science and Technology
75
,
1351
1361
.
Singh
R.
,
Das
T. K.
,
Kaur
R.
,
Raj
R.
&
Shekhawat
K.
2018
Weed management in dryland agriculture in India for enhanced resource use efficiency and livelihood security
.
Proceedings of the National Academy of Sciences India Section B – Biological Sciences
88
,
1309
1322
.
Standards, B. of I.
1988
Methods of Sampling and Test (Physical and Chemical) for Water and Wastewater. Part 32 Chloride (First Revision), Bureau of Indian Standards Is : 3025 (Part 32)-1988
.
Thakker
D.
2013
Jamnagar's Rangmati-Nagmati river now a garbage dump. Rajkot News-Times of India
.
Tomida
S.
,
Hanai
T.
,
Honda
H.
&
Kobayashi
T.
1999
Construction of COD simulation model for activated sludge process by recursive fuzzy neural network
.
Journal of Bioscience and Bioengineering
88
,
215
220
.
Verlicchi
P.
,
Al Aukidy
M.
,
Galletti
A.
,
Zambello
E.
,
Zanni
G.
&
Masotti
L.
2012
A project of reuse of reclaimed wastewater in the Po Valley, Italy: polishing sequence and cost-benefit analysis
.
Journal of Hydrology
432–433
,
127
136
.