The development of effluent removal prediction is crucial in providing a planning tool necessary for the future development and the construction of a septic sludge treatment plant (SSTP), especially in the developing countries. In order to investigate the expected functionality of the required standard, the prediction of the effluent quality, namely biological oxygen demand, chemical oxygen demand and total suspended solid of an SSTP was modelled using an artificial intelligence approach. In this paper, we adopt the clonal selection algorithm (CSA) to set up a prediction model, with a well-established method – namely the least-square support vector machine (LS-SVM) as a baseline model. The test results of the case study showed that the prediction of the CSA-based SSTP model worked well and provided model performance as satisfactory as the LS-SVM model. The CSA approach shows that fewer control and training parameters are required for model simulation as compared with the LS-SVM approach. The ability of a CSA approach in resolving limited data samples, non-linear sample function and multidimensional pattern recognition makes it a powerful tool in modelling the prediction of effluent removals in an SSTP.

INTRODUCTION

In developed countries, the centralization of wastewater treatment in urban areas is a usual practice. It contributes to a clean water supply and proper sanitation. Nevertheless, in the developing countries, using a septic sludge treatment plant (SSTP) through a sequence batch reactor (SBR) technology has been identified to ensure the removal of the effluent to the requirements of Standard A.

Based on the study by Huang et al. (2010), the efficient operations of a wastewater treatment process are limited because the process is affected by a variety of physical, chemical, and biological factors. The effectiveness of a treatment plant is based on the quality of the effluent being removed (Ye et al. 2009). However, lack of suitable process variables limits the effective control of the effluent quality (Harremoes et al. 1993). An SSTP is being modelled through a novel approach of a bio-inspired technique. The realism of the model can be proved based on the concept of biological systems.

The structure and the coefficients of the SSTP model consist of complex treatment processes, which are non-linear and characterized by many uncertainties within the influent parameters. The method used is named as a soft computing paradigm. It includes the clonal selection algorithm (CSA) and the least-square support vector machine (LS-SVM) approaches. In this study, both the immune-inspired and the support vector technique were developed and applied in an SSTP in Kuching City, Sarawak. The prediction of the effluent removals from an SSTP, such as biological oxygen demand (BOD), chemical oxygen demand (COD) and total suspended solids (TSS), was modelled using both applications. Monthly effluent samples were collected from the Matang SSTP in Kuching City, Sarawak, for the years 2007 to 2011 (60 samples) through an in-house laboratory. The calibration of the treatment plant used 5 years of monthly effluent data from the Matang SSTP using CSA and LS-SVM computational paradigm approach. As reported, a minimum of six samples per year were taken for wastewater treatment plant with a population equivalent of less than 2,000. Twelve samples must be taken in a year from 10,000 to 49,999 population equivalent wastewater treatment plants (Irish Department of the Environment and Local Government 2001). Alternatively, over-sampling may probably be a waste of time and resources as reported in EPA (2007).

SSTP MODEL DEVELOPMENT

As presented in the ‘Introduction’, an affordable wastewater disposal system such as the SSTP has been used. This has been practiced in Bangkok, Teheran (Orth 2007) and Sarawak (Ting et al. 2013, 2014). The SSTP is capable of treating septic sludge to the requirements of the Malaysia Environmental Quality Act 1974, Environmental Quality (Sewage) Regulation 2009 as Standard A. The cost of carrying out this treatment and the construction of a treatment plant is expensive. Thus, a thorough planning is required in order to identify the effectiveness of a treatment plant in compliance with the required Standard A of the Environmental Quality (Sewage) Regulation in the future. The prediction of the effluent removals with an SSTP model was investigated using an artificial intelligence approach.

ARTIFICIAL INTELLIGENCE

A computer simulation of an activated sludge process, called the Activated Sludge Model (ASM) 1 was developed by the International Association of Water Quality in 1987 (Henze et al. 1987). ASM 1 was developed to eliminate the removal of the organic carbon substances and nitrogen. Since then, other models have been proposed such as ASM 2 and ASM 2d which include the removal of phosphorus (Henze et al. 1995, 1999) and ASM 3 which helps to eradicate the carbon and nitrogen concurrently (Gujer et al. 1999).

Researchers further developed a steady-state model that includes the hydraulic model (Novak et al. 1999). In the study, Novak et al. (1999) developed full-scale data with regard to mass balance. However, the model depended entirely on dynamic data, which might encounter problems with the real input variables, which were usually faster than the slow process during the steady-state calibration. Thus, this model required frequent sampling for further accuracy (Petersen et al. 2002). Nevertheless, the measurements of a full-scale effluent removal were relatively expensive.

A model calibration for nitrogen removal and COD removal through laboratory-scale experiments was developed (Petersen et al. 2002). The full-scale model was required for more realistic results. The treatment of wastewater is a complex and non-linear biological reaction process. It is difficult to obtain reliable kinetic parameters and the prediction of a wastewater treatment may not be exact. However, intelligent algorithms are able to establish a complex system model (Hamoda et al. 1999; Hanbay et al. 2008). Since then, more soft computing techniques have been developed such as feed-forward neural network (Du et al. 2006), back propagation neural network (Emad et al. 2010) and recurrent high order neural network (Qiao et al. 2011). However, a direct cause–effect relationship to the wastewater treatment performance is rarely developed. The experimental results could lead to a conflicting result during modelling (Belanche et al. 2000).

The clonal selection principle of an artificial immune system (AIS) is being employed to identify the most effective predictive model. Although relatively young, the AIS has appeared to be an attractive field in modelling (Dasgupta 2006). Conversely, the SSTP is developed using the SVMs as a baseline model. The SVM is well recorded as a successful technique in prediction modelling (Daniel et al. 2013). Both algorithms were applied using MATLAB 7.10.

Clonal selection algorithm

Timmis (2007) stated that a large part of the work in AIS was based on clonal selection theory. The clonal selection theory of immunity states that an antigen being selected from among a variety of lymphocytes with receptors are capable of reacting with part of the antigen. CSA is among the most famous AIS techniques. It is designed based on the clonal selection principle. It has been verified that this technique has a great number of useful mechanisms from the viewpoint of programming, controlling, information processing and many more (Engin & Doyen 2004).

The CSA essentially centred on a repeated cycle of match, clone, mutate and replace (Greensmith et al. 2010). Numerous numbers of its parameters can be tuned, including the cloning rate, the initial number of antibodies, and the mutation rate for the clones. The proposed simulation of the SSTP model for SBR technology will undergo an initialization process where each antibody of real numbers, i.e. COD, TSS and BOD, represents the antibodies and a random variable of antigen in the immune system. On the model training, the affinity between the antibody and antigen is selected. In this study, the number of detectors selected is 200 for COD and TSS and 100 for BOD (Ting et al. 2012). The larger affinity will clone more offsprings in order to protect the good antibody and accelerate the convergent rate of the algorithm. In the cloning process, the best individual the cloned group will be selected to compose the next generation. This is used to evaluate and predict the performance of the SSTP in the long term.

Least-square support vector machine

The SVM is a branch of a neural network technique. SVM has resulted in promising pattern recognitions (Burges 1998). It is capable of solving small sample, non-linear, high dimension and local minimum points (Wang 2009). The kernel method is an important and efficient data mining technique dealing with large real-world datasets (Hofmann et al. 2008). The kernel method is introduced to deal with database inputs with the SVM. This will increase the computational efficiency and improve the accuracy of the prediction. An improved SVM with a radial basis function (RBF) kernel is employed in this SSTP model, called the LS-SVM. The LS-SVM is reported to be the most popular model which provides a better prediction result (Suykens & Vandewalle 1999). The LS-SVM method is applied as a baseline model. The LS-SVM method has been successfully developed and applied in the prediction of quality parameters in wastewater (Huang et al. 2009; Daniel et al. 2013).

The LS-SVM optimization process is utilized to map the data from an input space to a high-dimensional feature space before completing the linear regression. In order to decrease the calculation error caused by a variety of data sets, the sample data are normalized. Considering that the data used in the experiment are positive, the sample has been shrunk to [0,1]. The results obtained from the predictive simulation are anti-normalization data. In this study, RBF was used in the experiment as the Kernel function. The optimal parameters used were: the bias term associated with the Hyperplane, b = 64.0, and the error variable introduced in the sense of least-square minimization, e = 0.03125. This produced good prediction results in the sewage outflow simulation as proposed by Yang et al. (2011).

SIMULATION RESULTS

The training of sample data is the main factor that affects the learning and generalization abilities of the population. The training stage was carried out to obtain plots of the effluent discharge attributes' distributions of the SSTP. The model was trained with 36 effluent samples and the results are tabulated in Table 1. The accuracy of the training was between 91 and 98%, showing that the model was well trained.

Table 1

Training results of CSA versus LS-SVM approach

 Training results
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 91.69 91.73 
TSS 95.87 95.89 
BOD 97.97 98.16 
 Training results
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 91.69 91.73 
TSS 95.87 95.89 
BOD 97.97 98.16 

The SSTP model was further validated to identify the performance of the simulation. The number of samples changed from 36 samples on training to 12 samples during validation. The results show an increase in the results accuracy as tabulated in Table 2. The model validation has increased the accuracy from 91 to 94%. This finding also agrees that the model was well calibrated between the actual and predicted results. This signified the accurateness of the SSTP model through CSA and LS-SVM approach.

Table 2

Validation results of CSA versus LS-SVM approach

 Validation results
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 94.88 94.93 
TSS 98.79 98.79 
BOD 98.67 98.69 
 Validation results
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 94.88 94.93 
TSS 98.79 98.79 
BOD 98.67 98.69 

Results prediction

The results showed that the CSA approach was equally efficient and reliable as the LS-SVM approach as presented in Table 3. The comparison between actual and predicted results of CSA and LS-SVM methods presented a variation of 0.04% which was very minimal. The physical properties of effluent parameters such as COD and TSS have significant environmental implications. Both approaches have shown the successfulness of the algorithm to predict the effluent discharge with an accuracy of 94 and 98%. The BOD, described as the amount of oxygen required by microorganisms to break down organic matter, has good prediction with an accuracy of 98%. The theory of immunity as mentioned in the ‘Artificial intelligence’ section has indicated the relationship between the biological processes of a treatment plant can be analogous to the attraction of lymphocytes with receptors. In short, the CSA approach demonstrates an alternative method in the prediction of the effluent removals in an SSTP.

Table 3

Efficiency of CSA versus LS-SVM approach

 Prediction method
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 94.09 94.07 
TSS 98.74 98.74 
BOD 98.46 98.41 
 Prediction method
 
Effluent discharge CSA (%) LS-SVM (%) 
COD 94.09 94.07 
TSS 98.74 98.74 
BOD 98.46 98.41 
The model's efficiency is calculated using the following formula: 
formula
1
where TN is the test result of true negative; FP is the test result of false positive; FN is the test result of false negative; TP is the test result of true positive.

Performance fitness function

The results recorded in Table 3 were tested for the goodness of fit. Since the LS-SVM approach was used as a baseline model, the developed CSA approach was tested statistically against the LS-SVM approach. The results showed that the TSS (Figure 1), COD (Figure 2) and BOD (Figure 3) had an R2-value of 1.000 which indicates the efficiency of the model's effluent real data and the simulated results. The mean absolute percentage error (MAPE) presented the results of 0.0109, 0.0511 and 0.0511 for TSS, COD and BOD effluent removals, respectively. Root mean square error (RSME) values of 0.2 for TSS, and 0.7483 for COD and BOD were obtained, respectively. These performance indexes proved that the CSA method was equally excellent as the LS-SVM method for the prediction purposes.

Figure 1

Test performance obtained from testing of TSS effluent between CSA and LS-SVM methods.

Figure 1

Test performance obtained from testing of TSS effluent between CSA and LS-SVM methods.

Figure 2

Test performance obtained from testing of COD effluent between CSA and LS-SVM methods.

Figure 2

Test performance obtained from testing of COD effluent between CSA and LS-SVM methods.

Figure 3

Test performance obtained from testing of BOD effluent between CSA and LS-SVM methods.

Figure 3

Test performance obtained from testing of BOD effluent between CSA and LS-SVM methods.

CONCLUSIONS

In this study, a new method of predicting the effluent removals of BOD, COD and TSS using a CSA method has been proposed. Furthermore, the data obtained from an actual full-scale SSTP were used to verify the validity of the proposed model. Based on the findings, the developed CSA-based SSTP model could be used to determine the performance of a treatment plant and the prediction of the effluent removals. As compared with the LS-SVM method, the CSA method is correspondingly more effective in the prediction of the effluent discharge from an SSTP.

REFERENCES

REFERENCES
Belanche
L.
Valdes
J. J.
Comas
J.
Roda
I. R.
Poch
M.
2000
Prediction of the bulking phenomenon in wastewater treatment plants
.
Artificial Intelligence in Engineering
14
,
307
317
.
Burges
C. J. C.
1998
A tutorial on support vector machines for pattern recognition
.
Data Mining Knowledge Discovery
2
,
121
167
.
Daniel
R.
Sanfins
A.
Belo
O.
2013
Wastewater treatment plant performance prediction with support vector machines
. In:
ICDM 2013, LNAI 7987
, (
Perner
P.
, ed.).
Springer
,
Berlin
, pp.
99
111
.
Dasgupta
D.
2006
Advance in artificial immune system
.
IEEE Computational Intelligence Magazine
1
(
4
),
40
49
.
Du
H. B.
Shao
H. H.
Yao
P. J.
2006
Adaptive neural network control for a class of low-triangular-structured nonlinear systems
.
IEEE Transactions on Neural Networks
17
(
2
),
509
514
.
Emad
S. E.
Chaudhuri
M.
Mohamed
M. E.
2010
The use of artificial neural network (ANN) for modeling of COD removal from antibiotic aqueous solution by the Fenton process
.
Journal of Hazardous Materials
179
(
1–3
),
127
134
.
EPA (Environment Protection Authority)
2007
Regulatory monitoring and testing water and wastewater sampling. EPA, Adelaide, Australia. http://www.epa.sa.gov.au/xstd_files/Water/Guideline/guide_wws.pdf.
Greensmith
J.
Whitbrook
A.
Aickelin
U.
2010
Handbook of Metaheuristics
. 2nd edn,
Chapter 14 (Artificial Immune Systems)
,
Kluwer Academic, Norwell, MA, USA
, pp.
421
448
.
Gujer
W.
Henze
M.
Mino
T.
van Loosdrecht
M. C. M.
1999
Activated sludge model No. 3
.
Water Science Technology
39
(
1
),
183
193
.
Hamoda
M. F.
Ghusain
I. A.
Hassan
A. H.
1999
Integrated wastewater treatment plant performance evaluation using artificial neural networks
.
Water Science and Technology
40
,
55
65
.
Harremoes
P.
Capodaglio
A. G.
Hellstrom
B. G. M.
Henze
K. N.
Jensen
A.
Lynggaaard-Jensen
R.
Otterpohl
H.
1993
Wastewater treatment plants under transient loading performance, modeling and control
.
Water Science and Technology
27
(
12
),
71
115
.
Henze
M.
Grady
C. P. L.
Jr
Gujer
W.
Marais
G. R.
Matsuo
T.
1987
Activated Sludge Model No. 1, IAWPRC Scientific and Technical Reports 1
.
IAWPRC
,
London
.
Henze
M.
Gujer
W.
Mino
T.
Matsuo
T.
Wentzel
M. C.
Marais
G. v. R.
1995
Activated Sludge Model No. 2, IAWPRC Scientific and Technical Reports 3
.
IAWPRC
,
London
.
Henze
M.
Gujer
W.
Mino
T.
Matsuo
T.
Wentzel
M. C.
Marais
G. v. R.
van Loosdrecht
M. C. M.
1999
Activated sludge model No. 2d, ASM 2d
.
Water Science Technology
39
(
1
),
165
182
.
Hofmann
T.
Scholkopf
B.
Smola
A. J.
2008
Kernel methods in machine learning
.
The Annals of Statistics
36
(
3
),
1171
1220
.
Huang
Z.
Luo
J.
Li
X.
Zhou
Y.
2009
Prediction of effluent parameters of wastewater treatment plant based on improved least square support vector machine with PSO
. In:
Proceeding 1st International Conference on Information Science and Engineering, ICISE (2009)
.
Irish Department of the Environment and Local Government
2001
Urban Waste Water Treatment Regulations, 2001. S.I. No. 254/2001. http://www.irishstatutebook.ie/2001/en/si/0254.html (accessed 18 November 2014)
.
Novak
O.
Franz
A.
Svardal
K.
Muller
V.
Kuhn
V.
1999
Parameter estimation for activated sludge models with help of mass balances
.
Water Science and Technology
39
(
4
),
113
120
.
Orth
H.
2007
Centralised versus decentralised wastewater systems? In:
Proceedings GWA Gewasserschutz-Wasser-Abwasser, Band 206, Advanced Sanitation Conference, International IWA Conference, 12–13 Mac, Aachen, Germany
, pp.
421
428
.
Petersen
B.
Gernaey
K.
Henze
M.
Vanrolleghem
P. A.
2002
Evaluation of an ASM 1 model calibration procedure on municipal-industrial wastewater treatment plant
.
Journal of Hydroinformatics
4
, pp.
15
38
.
Qiao
J. F.
Yang
W. W.
Yuan
M. Z.
2011
Recurrent high order neural network modeling for wastewater treatment process
.
Journal of Computers
6
(
8
),
1570
1577
.
Suykens
J. A.
Vandewalle
J.
1999
Least squares support vector machine classifiers
.
Neural Processing Letters
9
(
3
),
293
300
.
Ting
S. C.
Ismail
A. R.
Malek
M. A.
2012
Bio-engineered solution in enhancement of septic sludge with artificial immune system
. In:
Proceedings of the 2nd International Conference on Water Resources (ICWR 2012), 5–6 November, 2012, Langkawi, Kedah
.
Ting
S. C.
Malek
M. A.
Ismail
A. R.
2014
Prediction analysis of effluent removal in septic sludge treatment plant: a biomimetics engineering approach
.
Journal of Environmental Science Processes & Impacts
16
,
2208
2214
.
Wang
D. C.
2009
Support Vector Machines Modeling Prediction and Control
.
China Meteorological
,
Beijing
.
Yang
B.
Zhao
D.
Zhang
J.
2011
Prediction system of sewage outflow COD based on LS-SVM
. In:
Proceedings of the 2nd International Conference on Intelligent Control and Information Processing, 25–28 July
.
Ye
H.
Luo
F.
Xu
Y.
2009
Application of RBF Network Based on Immune Algorithm to Predicting of Wastewater Treatment
.
Springer-Verlag
,
Berlin
, pp.
1197
1202
.