Raw water quality variation has a great effect on drinking water treatment. To improve the adaptivity of drinking water treatment and stabilize the quality of treated water, a raw water quality assessment method, which is based upon the support vector machine (SVM), is developed in this study. Compared to existing raw water quality assessment methods, the assessment method studied herein is oriented to drinking water treatment and can directly be used for the control of the chemical (alum and ozone) dosing process. To this end, based upon the productive experiences and the analysis of the operating data of water supply, a raw water quality assessment standard oriented to drinking water treatment is proposed. A raw water quality model is set up to assess the raw water quality based upon the SVM technique. Based upon the raw water quality assessment results, a feedforward–feedback control scheme has been designed for the chemical dosing process control of drinking water treatment. Thus, the chemical dosage can be adjusted in time to cope with raw water quality variations and hence, the quality of the treated water is stabilized. Experimental results demonstrate the improved effectiveness of the proposed method of raw water quality assessment and the feedforward–feedback control scheme.

INTRODUCTION

With the pollution of water by natural and anthropogenic activities, the safety of drinking water and people's health face an imminent threat. For this reason, the quality of drinking water sources has now become a matter of serious concern (Jurado et al. 2012; Liu et al. 2014; Ribeiro et al. 2015). Drinking water is produced by a series of water treatment processes, such as pre-ozonation, coagulation, flocculation, sedimentation, sand filtration, main ozonation, biological activated carbon (BAC) filtration and chlorination, which are severely affected by variations in raw water quality (Wenk et al. 2013; Sallanko & Väisänen 2013). Therefore, it is very important to acquire real-time information of raw water quality for drinking water treatment. There are many water quality parameters for describing the characteristics of raw water (Benson 2006; Hamzah 2007). However, the online measurement of all water quality parameters is impossible due to financial and technological constraints. As a result, some water quality parameters such as temperature, turbidity, chemical oxygen demand (CODMn), ammoniacal nitrogen (NH4-N) and pH are monitored in real-time by online instruments, while other water quality parameters such as dissolved organic carbon, total nitrogen (TN) and total phosphorus (TP) are measured by laboratory analysis (on a daily, weekly or monthly basis) (Parashar et al. 2008; Ghumman 2011).

Assessment of raw water quality is commonly developed to classify the raw water quality of natural waters and is determined using certain mathematical methods (Ross 1977; Vega et al. 1998; Singh et al. 2005; Zhou et al. 2007; Chu et al. 2013). To the best of our knowledge, reports on the assessment of raw water quality for the treatment of drinking water are non-existent in the literature. One reason for this is that the safety of drinking water in many countries is not taken enough into account despite poor raw water quality and significant raw water quality changes (Schulz et al. 1992; Gadgil 1998; Sobsey et al. 2008). On the other hand, drinking water treatment is not only related to the water quality grades of natural waters, but is also related to factors such as climate, weather and discharge of wastewater (Delpla et al. 2011; Stepien et al. 2014). Thus, these assessment methods can only be used for the off-line support and cannot be used directly for the real-time control of drinking water treatment. The raw water quality of the water treatment plant studied in this work has been getting worse over recent years, and the changes in raw water quality have been increasingly frequent, especially in heavy rains and strong winds. This is because the raw water intake site of the water treatment plant studied in this work is located in shallow waters. High turbidity is often seen after heavy rains and strong winds affect the shallow lake bottom (Zhu et al. 2005, 2007). Therefore, research on raw water quality assessment oriented to drinking water treatment is urgently needed to stabilize the quality of the treated water.

The support vector machine (SVM), originally proposed by Vapnik (1995), is a powerful machine-learning algorithm based on statistical learning theory. SVM applies the structural risk minimization (SRM) principle and possesses better generalization ability (Wang & Fu 2005). Until now, SVM has widely been used in various nonlinear problems for the purpose of classification, soft sensor, signal processing and so on (Mehta & Lingayat 2007; Liu et al. 2010; Pal & Foody 2010). In SVM, the parameters have great influence on the modeling accuracy and generalization ability. Traditional parameter determination methods such as trial and error, grid-search and Bayesian are based on estimation and approximation, and have the disadvantage of low precision and large calculations (Le-peng et al. 2006; Kalyani et al. 2011). Particle swarm optimization (PSO) is a swarm intelligence meta-heuristic behavior of decentralized systems obtained from the simulation of flocking birds or schooling fish. It carries on an intelligent search for the solution space through a ‘cooperative’ strategy of individuals (called particles), whereas a genetic algorithm (GA) is based on a ‘competitive’ strategy. Suboptimal solutions in the PSO algorithm can therefore survive and contribute to the search process at later stages of iteration. PSO has been proved to have better parameter optimization performance than GA by some researchers (Ren & Bai 2010; Tan et al. 2012).

The major contribution of this work is the development of a raw water quality assessment method, which can be applied to the real-time control of drinking water treatment. We have studied the historical data and productive experiences, and have established a drinking water treatment-oriented raw water quality assessment standard. The raw water quality assessment model is set up to assess raw water quality based on a PSO trained SVM modeling technique. The other novelty of this work is that the author has proposed a feedforward–feedback control scheme, which is preferable to the general control scheme of proportion–integration–differentiation (PID) feedback control for the chemical dosing process. To date, there have been successful experiments of the work presented in this manuscript in the practical process control system of Xiangcheng Water Treatment Plant (XWTP) in Suzhou, China.

The rest of this paper is organized as follows. The raw water quality assessment standard oriented to drinking water treatment is presented in the second section. The SVM modeling process is described in the third section. After a brief description of the feedforward–feedback control scheme of the chemical dosing process, experiments are conducted in the fourth section. Finally, conclusions are given in the fifth section.

RAW WATER QUALITY ASSESSMENT STANDARD

Raw water quality analysis

The raw water of XWTP is obtained from Jinsu Bay of Taihu Lake, which is one of the five biggest freshwater lakes in China. The seasonal variation of the raw water quality of the water source area is significant. The contents of TN and TP in the water are higher in summer, whereas the concentration of ammoniacal nitrogen is higher in winter. In addition, the turbidity is significantly affected by the weather, especially the influence of heavy rain and strong wind on the shallow lake bottom. The ranges for raw water quality parameters in XWTP from January 2012 to December 2013 are presented in Table 1.

Table 1

Raw water quality in XWTP from January 2012 to December 2013

ParameterMaxMinAverage
pH 9.3 6.9 8.4 
Temperature/°C 33.9 1.5 16.7 
Turbidity/NTU 197.4 9.2 25.7 
NH4+-N/mg L−1 2.21 0.003 0.22 
CODMn/mg L−1 5.2 1.7 2.8 
TP/mg L−1 0.527 0.003 0.052 
TN/mg L−1 4.62 0.079 1.29 
Bromide/mg L−1 0.378 0.182 0.272 
TOC/mg L−1 6.27 3.58 4.33 
ParameterMaxMinAverage
pH 9.3 6.9 8.4 
Temperature/°C 33.9 1.5 16.7 
Turbidity/NTU 197.4 9.2 25.7 
NH4+-N/mg L−1 2.21 0.003 0.22 
CODMn/mg L−1 5.2 1.7 2.8 
TP/mg L−1 0.527 0.003 0.052 
TN/mg L−1 4.62 0.079 1.29 
Bromide/mg L−1 0.378 0.182 0.272 
TOC/mg L−1 6.27 3.58 4.33 

Raw water quality assessment standard

To evaluate the influence of the variation of water quality parameters on the chemical (alum and ozone) dosing process, we studied the process operating data of chemical dosage under different conditions of raw water quality. Moreover, we referred to the suggestion of operators with rich experience, and chose temperature, turbidity, CODMn and NH4-N as the affecting factors of raw water quality assessment. Thus, the raw water quality assessment standard oriented to the control of the chemical dosing process is established based on the operating data analysis and operators' experience in Table 2.

Table 2

Raw water quality assessment standard

Water quality gradeTemperature/°CTurbidity/NTUNH4+-N/mg L−1CODMn/mg L−1Desired outputOptimum alum dosage/mg L−1Optimum ozone dosage/mg L−1
10 20 0.5 0.1 0.7 
II 10 40 0.2 11 0.8 
III 10 80 2.5 0.3 14 0.9 
IV 20 20 0.5 2.5 0.4 0.8 
20 40 0.5 10 0.9 
VI 20 80 1.5 3.5 0.6 13 
VII 30 20 0.5 0.7 0.9 
VIII 30 40 4.5 0.8 
IX 30 80 0.9 12 1.2 
Water quality gradeTemperature/°CTurbidity/NTUNH4+-N/mg L−1CODMn/mg L−1Desired outputOptimum alum dosage/mg L−1Optimum ozone dosage/mg L−1
10 20 0.5 0.1 0.7 
II 10 40 0.2 11 0.8 
III 10 80 2.5 0.3 14 0.9 
IV 20 20 0.5 2.5 0.4 0.8 
20 40 0.5 10 0.9 
VI 20 80 1.5 3.5 0.6 13 
VII 30 20 0.5 0.7 0.9 
VIII 30 40 4.5 0.8 
IX 30 80 0.9 12 1.2 

RAW WATER QUALITY ASSESSMENT MODELING

Raw water quality assessment modeling is crucial for the study of raw water quality. In this study, an SVM model is established for the real-time assessment of raw water quality, while a PSO algorithm has been used to train the SVM model parameters.

SVM model

The basic idea of SVM for regression is to trap the input data set into a high-dimensional feature space via nonlinear mapping and to solve the linear regression problem in this feature space (Dormido-Canto et al. 2004). An optimum decision function is constructed based on the principle of SRM. The structure of the SVM model is shown in Figure 1.
Figure 1

Schematic structure of the SVM model.

Figure 1

Schematic structure of the SVM model.

Consider a set of training data (X1,y1), …,(Xi,yi), …(Xl,yl) ∈ Rn × R, where Xi denotes the input vector, yi denotes the corresponding output value, and l denotes the total number of data patterns. The nonlinear mapping function (·) maps the original input space Rn to the higher dimensional feature space Rk: (X) = ((X1), …(Xl)), where k(k > >n) denotes the dimension of the feature space. The SVM regression function is formulated as 
formula
1
where is the weight vector, and b is the bias term. The coefficients and b can be estimated by minimizing the regularized risk function. By the SRM principle, we obtain the optimization problem: 
formula
2
where is the regularization term, controlling the complexity of the model. is the error term. C is the regularization parameter.
The solution to this optimization problem is given by the Lagrangian method. Thus, the nonlinear function of Equation (1) can be expressed as: 
formula
3
where are the Lagrangian multipliers.

We define kernels . Different kernels can construct different types of SVM (Bray & Han 2004; Khajeh & Modarress 2011). Three kernel functions have been commonly used in SVM:

  • (1) Polynomial kernel: .

  • (2) Sigmoid kernel: K(X, Xi) = tanh(kXXi + v), k > 0, .

  • (3) Radial basic function (RBF) kernel: K(X, Xi) = exp.

RBF is the most frequently used kernel, because it requires only one parameter to be determined, whereas the polynomial basis functions or sigmoid function require two parameters to be determined.

The SVM constructed from the RBF kernel has a similar overall performance to the SVM constructed from the other two kernels (Scholkopf et al. 1995; Lin et al. 2008). Therefore, RBF has been used as the preferred choice of kernel in this study.

Proper setting of parameters strongly affects the performance of the SVM model. In SVM with the RBF kernel, C and are the two major parameters. If the regularization parameter C is too large, the estimation accuracy rate becomes very low in the testing phase (though it remains very high in the training phase). If the regularization parameter C is too small, the estimation accuracy rate turns out to be unsatisfied, and the model becomes useless (Ding & Chen 2010). Kernel parameter determines the RBF kernel width and is related to the input range of the training data set. A very large value of kernel parameter will result in over-fitting, while a very small value will lead to under-fitting.

The PSO algorithm is a global optimization technology based on group intelligence, which carries on the intelligent search for the solution space through mutual effect in order to discover the optimum global solution (Kannan et al. 2004). Particle swarms explore the search space through a population of particles. During each iteration, every particle tries to find the best global solution by adjusting the trajectory towards its own best position and the best particle of the swarm . The velocity and position are updated according to the following equations: 
formula
4
 
formula
5
where i = 1,2, …,n, n is the number of the particle, is the present velocity of particle i, is the present position of particle i, k is the inertia number, and are the acceleration constants, and are random numbers selected in the range [0,1], is the inertia weight, which can be described as follows: 
formula
6
where is the max inertia number, and are the initial inertia weight and final inertia weight, respectively.
The fitness function of a particle is shown in the following equation: 
formula
7
where is the number of training samples, is the ideal output and is the actual output.
The training process of the SVM model parameters with PSO is shown in Figure 2.
Figure 2

Training process of SVM model parameters with PSO.

Figure 2

Training process of SVM model parameters with PSO.

Modeling results

We have analyzed the historical data in XWTP from January 2012 to December 2013 and have selected 300 groups of historical data. Every group of historical data represents the operating analysis made on one typical sample. The 300 groups of historical data are divided into two parts.

The first 200 groups are used for training the SVM model, while the remaining 100 groups are used for testing the performance of the SVM model. The convergence of the training process has been controlled by the root-mean-square error (RMSE) between the model output and the desired output. The RMSE is defined as 
formula
8
where is the model output, is the desired output and n is the sample number.
The parameters of the SVM model, namely C and , which yield the highest accuracy, are ultimately set to 3,000 and 0.16 separately. Figure 3 shows the training result of the SVM model and Figure 4 shows the testing result of the SVM model. It can be seen that the SVM model performs well in predicting the relevant required output.
Figure 3

Training result of the SVM model.

Figure 3

Training result of the SVM model.

Figure 4

Testing result of the SVM model.

Figure 4

Testing result of the SVM model.

The RMSE of the real-time assessment results of raw water quality for the testing periods of 18 April 2014 in XWTP is 0.035, which also verifies that the SVM model gives an acceptable result and can be applied to assess raw water quality.

CHEMICAL DOSING PROCESS CONTROL OF DRINKING WATER TREATMENT

The XWTP (capacity of 300,000 m3/day) was originally put into service in 2007. The overall drinking water treatment process of XWTP comprises pre-ozonation, coagulation, flocculation, sedimentation, sand filtration, main ozonation, BAC filtration and chlorination. The process is shown schematically in Figure 5. The alum dosing process before the sedimentation tank and the ozone dosing process before the main ozone contactor are crucial for the overall drinking water treatment. These directly affect the quality of the treated water.
Figure 5

Flow diagram of the drinking water treatment process.

Figure 5

Flow diagram of the drinking water treatment process.

Feedforward–feedback control scheme

Like most drinking water treatment plants of China, the chemical (alum and ozone) dosing process in XWTP adopts the PID feedback control. In PID feedback control, the control action is calculated according to proportion, integration and differentiation on the basis of deviation between the actual output and the set-point (Cromphout et al. 2005; Lian et al. 2012). Owing to the complexity and reactivity of the substances involved, both the chemical dosing processes during drinking water treatment and the large time-delays are seriously affected by various factors, such as water flow, water quality, and chemical dosage. It is difficult to stabilize the quality of treated water for the traditional PID feedback control scheme, especially during a period of major changes in raw water quality. In this paper, we present a feedforward–feedback control scheme as shown in Figure 6. With the feedforward–feedback control scheme, a feedforward compensation, which is based upon the aforementioned raw water quality assessment, has been designed.
Figure 6

Feedforward–feedback control scheme for the chemical dosing process.

Figure 6

Feedforward–feedback control scheme for the chemical dosing process.

Experimental results

In order to test the effects of the proposed feedforward–feedback control scheme based on raw water quality assessment, experiments have been made on the raw water quality assessment method together with the feedforward–feedback control scheme at the XWTP. The assessment model and the control scheme are coded in the commercial SCADA software of Siemens Wincc. All the online signals from or to the chemical dosing process control system are interconnected through a distributed control system (DCS) as shown in Figure 7. Process data are saved in a database in a PC server. The control scheme is programmed on a PC and is executed through a programmable logic controller.
Figure 7

Distributed control system (DCS) for the chemical dosing process.

Figure 7

Distributed control system (DCS) for the chemical dosing process.

The integral of absolute error (IAE) is chosen as the quantitative index to evaluate the control performance: 
formula
9
where is the reference signal, is the actual process output.
The experimental results of the proposed feedforward–feedback control scheme and PID feedback control scheme are shown in Figure 8 and Table 3. The parameters of the PID controller are adjusted to the best control effect by using the critical ratio method, which is used frequently in engineering applications. The corresponding assessment results of raw water quality are listed in Table 4. The percentage reduction of total organic carbon (TOC)/turbidity is presented in Table 5.
Table 3

Performance index of experimental results of chemical dosing process control

MethodControl objectiveIAE (mg L−1)
PID feedback control Turbidity 0.19 
ResO3 0.055 
Feedforward–feedback control Turbidity 0.08 
ResO3 0.02 
MethodControl objectiveIAE (mg L−1)
PID feedback control Turbidity 0.19 
ResO3 0.055 
Feedforward–feedback control Turbidity 0.08 
ResO3 0.02 
Table 4

Assessment results of raw water quality for Figure 8 

Time/hourTemperature/°CTurbidity/NTUNH4+-N/mg L−1CODMn/mg L−1Assessment outputOptimum alum dosage/mg L−1Optimum ozone dosage/mg L−1Water quality grade
(a) PID feedback control 
 0 21.2 23 0.65 2.6 0.43 8.6 0.83 IV 
 1 22.7 46 1.22 3.1 0.53 10.9 0.93 
 2 22.8 70 1.32 3.6 0.61 12.4 0.99 VI 
 3 22.2 77 1.43 3.7 0.64 10.6 0.96 VI 
 4 22.9 62 1.12 3.1 0.53 10.9 0.93 
 5 21.7 82 1.52 3.4 0.59 12.7 0.99 VI 
 6 22.5 69 1.25 3.8 0.62 11.8 0.98 VI 
 7 22.2 52 1.11 3.7 0.67 8.8 0.93 VII 
(b) Feedforward–feedback control 
 0 16.2 31 1.15 3.2 0.51 10.3 0.91 
 1 16.9 41 1.22 3.5 0.55 11.5 0.95 
 2 17.3 58 1.35 3.4 0.59 12.7 0.99 VI 
 3 17.2 65 1.62 3.8 0.68 8.2 0.92 VII 
 4 17.6 66 1.59 3.3 0.68 8.2 0.92 VII 
 5 16.9 70 1.44 3.6 0.63 11.2 0.97 VI 
 6 16.5 72 1.37 3.2 0.63 11.2 0.97 VI 
 7 17.2 75 1.28 3.3 0.57 12.1 0.97 VI 
Time/hourTemperature/°CTurbidity/NTUNH4+-N/mg L−1CODMn/mg L−1Assessment outputOptimum alum dosage/mg L−1Optimum ozone dosage/mg L−1Water quality grade
(a) PID feedback control 
 0 21.2 23 0.65 2.6 0.43 8.6 0.83 IV 
 1 22.7 46 1.22 3.1 0.53 10.9 0.93 
 2 22.8 70 1.32 3.6 0.61 12.4 0.99 VI 
 3 22.2 77 1.43 3.7 0.64 10.6 0.96 VI 
 4 22.9 62 1.12 3.1 0.53 10.9 0.93 
 5 21.7 82 1.52 3.4 0.59 12.7 0.99 VI 
 6 22.5 69 1.25 3.8 0.62 11.8 0.98 VI 
 7 22.2 52 1.11 3.7 0.67 8.8 0.93 VII 
(b) Feedforward–feedback control 
 0 16.2 31 1.15 3.2 0.51 10.3 0.91 
 1 16.9 41 1.22 3.5 0.55 11.5 0.95 
 2 17.3 58 1.35 3.4 0.59 12.7 0.99 VI 
 3 17.2 65 1.62 3.8 0.68 8.2 0.92 VII 
 4 17.6 66 1.59 3.3 0.68 8.2 0.92 VII 
 5 16.9 70 1.44 3.6 0.63 11.2 0.97 VI 
 6 16.5 72 1.37 3.2 0.63 11.2 0.97 VI 
 7 17.2 75 1.28 3.3 0.57 12.1 0.97 VI 
Table 5

Percentage reduction of TOC/turbidity

Control methodTOCTurbidity
PID feedback control 34.2% 89.2% 
Feedforward–feedback control 37.8% 93.3% 
Control methodTOCTurbidity
PID feedback control 34.2% 89.2% 
Feedforward–feedback control 37.8% 93.3% 
Figure 8

Experimental results of chemical dosing process control: (a) PID feedback control; (b) feedforward–feedback control.

Figure 8

Experimental results of chemical dosing process control: (a) PID feedback control; (b) feedforward–feedback control.

It can be seen from the experimental results (Figure 8 and Table 3) that the proposed feedforward–feedback control scheme based on raw water quality assessment provides a better control performance than the PID feedback control scheme. This is consistent with the previous theoretical analysis, which has shown that the feedforward–feedback control scheme can adjust the chemical dosage in time to cope with the variation of raw water quality. Thus, more stable treated water quality is provided and the consumer's health is protected from chemical and microbiological risks, especially during periods of frequent changes in raw water quality. At the same time, the proposed feedforward–feedback control scheme makes the operation of drinking water treatment more efficient and brings great improvements in plant management.

CONCLUSIONS

Raw water quality assessment oriented to drinking water treatment has been conducted in this paper. Based upon the productive experiences and analysis of the operating data of a drinking water treatment plant, a raw water quality assessment standard for the process control of drinking water treatment has been put forward. An SVM model trained by a PSO algorithm is used to establish the real-time assessment model of raw water quality. A feedforward–feedback control scheme based on the raw water quality assessment is designed for the chemical dosing process control of drinking water treatment to cope with raw water quality variations and to stabilize the quality of treated water. Experimental results showed an increase in the performance of the control scheme compared to the previous PID feedback scheme.

ACKNOWLEDGEMENTS

This work was supported by Natural Science Foundation of Jiangsu Province (Grant No. BK20150841), NUPTSF (Grant No. NY214019) and NUPTSF (Grant No. NY214078).

REFERENCES

REFERENCES
Benson
Y.
2006
Effect of Raw Water Quality on Coagulant Dosage and Optimum pH. Masters thesis, Faculty of Civil Engineering, Universiti Teknologi Malaysia
.
Bray
M.
Han
D.
2004
Identification of support vector machines for runoff modelling
.
Journal of Hydroinformatics
6
,
265
280
.
Chu
H. B.
Lu
W. X.
Zhang
L.
2013
Application of artificial neural network in environmental water quality assessment
.
Journal of Agricultural Science and Technology
15
(
2
),
343
356
.
Cromphout
J.
Walravens
E.
Vanhoucke
R.
2005
Improvement of water quality in the drinking water plant of Kluizen by the use of ozone in combination with GAC
.
Tribune de l'Eau
58
(
633
),
15
18
.
Delpla
I.
Baures
E.
Jung
A. V.
Clement
M.
Thomas
O.
2011
Issues of drinking water quality of small scale water services towards climate change
.
Water Science and Technology
63
(
2
),
227
232
.
Dormido-Canto
S.
Farias
G.
Dormido
R.
2004
TJ-II wave forms analysis with wavelets and support vector machines
.
Review of Scientific Instruments
75
(
10
),
4254
4257
.
Gadgil
A.
1998
Drinking water in developing countries
.
Annu. Rev. Energy Environ.
23
,
253
286
.
Ghumman
A. R.
2011
Assessment of water quality of Rawal Lake by long-time monitoring
.
Environmental Monitoring and Assessment
180
(
1–4
),
115
126
.
Hamzah
N.
2007
Assessment on Water Quality and Biodiversity within Sungai Batu Pahat. Masters thesis, Faculty of Civil Engineering, Universiti Teknologi Malaysia
.
Jurado
A.
Vazquez–Sune
E.
Carrera
J.
Lopez de Alda
M.
Pujades
E.
Barcelo
D.
2012
Emerging organic contaminants in groundwater in Spain: a review of sources, recent occurrence and fate in a European context
.
Science of the Total Environment
440
,
82
94
.
Kalyani
S.
Swarup
K. S.
Kalyani
S.
2011
Classification and assessment of power system security using multiclass SVM
.
IEEE Transactions on Systems, Man, and Cybernetics Part C (Applications and Reviews)
41
(
5
),
753
758
.
Kannan
S.
Mary Raja Slochannal
S.
Subbaraj
P.
Narayana Prasad
P.
2004
Application of particle swarm optimization technique and its variants to generation expansion planning problem
.
Electric Power Systems Research
70
,
203
210
.
Khajeh
A.
Modarress
H.
2011
Quantitative structure–property relationship for flash points of alcohols
.
Industrial & Engineering Chemistry Research
50
(
19
),
11337
11342
.
Le-peng
B.
Wei
X.
Han-tao
S.
2006
A Bayesian classification algorithm based on one-class SVM
.
Transactions of Beijing Institute of Technology
26
(
2
),
143
146
.
Lian
Z.
Liu
G.
Pan
L.
2012
PID Algorithm applied in the refrigeration of EMBCCD
.
Optical Technique
552
(
6
),
459
462
.
Lin
S. W.
Ying
K. C.
Chen
S. C.
2008
Particle swarm optimization for parameter determination and feature selection of support vector machines
.
Expert Systems with Applications
35
(
4
),
1817
1824
.
Liu
G.
Zhou
D.
Xu
H.
2010
Model optimization of SVM for a fermentation soft sensor
.
Expert Systems with Applications
37
(
4
),
2708
2713
.
Liu
Z. Q.
Han
B. J.
Wen
G.
Ma
J.
Wang
S. J.
Zha
R. G.
Shen
L. P.
Wang
C.
2014
Full-scale application of catalytic ozonation for drinking water treatment: case study in China
.
Journal of Environmental Engineering
140
(
9
),
1
8
.
Mehta
S. S.
Lingayat
N. S.
2007
Biomedical signal processing using SVM. In: IET-UK International Conference on Information and Communication Technology in Electrical Sciences (ICTES 2007)
, pp.
527
532
.
Pal
M.
Foody
G. M.
2010
Feature selection for classification of hyperspectral data by SVM
.
IEEE Transactions on Geoscience and Remote Sensing
48
(
5
),
2297
2307
.
Parashar
C.
Verma
N.
Dixit
S.
Shrivastava
R.
2008
Multivariate analysis of drinking water quality parameters in Bhopal, India
.
Environmental Monitoring and Assessment
140
(
1–3
),
119
122
.
Ren
Y.
Bai
G.
2010
Determination of optimal SVM parameters by using GA/PSO
.
Journal of Computers
5
(
8
),
1160
1168
.
Ross
S. L.
1977
An index system for classifying river water quality
.
Water Pollution Control
76
(
1
),
113
122
.
Sallanko
J.
Väisänen
T.
2013
Effects of ozonation on AOC content of humic Finnish groundwater
.
Ozone: Science & Engineering
35
(
2
),
86
89
.
Scholkopf
B.
Burges
C.
Vapnik
V.
1995
Extracting support data for a given task
. In:
Proceedings, First International Conference on Knowledge Discovery & Data Mining
.
AAAI Press
,
Menlo Park, CA
, pp.
252
257
.
Schulz
C. R.
Okun
D. A.
Donaldson
D.
Austin
J.
1992
Surface Water Treatment for Communities in Developing Countries
.
John Wiley & Sons
,
New York
,
USA
.
Sobsey
M. D.
Stauber
C. E.
Casanova
L. M.
Brown
J. M.
Elliott
M. A.
2008
Point of use household drinking water filtration: a practical, effective solution for providing sustained access to safe drinking water in the developing word
.
Environmental Science & Technology
42
(
12
),
4261
4267
.
Stepien
D. K.
Diehl
P.
Helm
J.
Thorns
A.
Püttmann
W.
2014
Fate of 1,4-dioxane in the aquatic environment: from sewage to drinking water
.
Water Research
48
(
1
),
406
419
.
Tan
J.
Chen
X.
Du
M.
2012
An internet traffic identification approach based on GA and PSO-SVM
.
Journal of Computers
7
(
1
),
19
29
.
Vapnik
V.
1995
The Nature of Statistical Learning Theory
.
Springer-Verlag
,
Berlin, Heidelberg, and New York
.
Wang
L. P.
Fu
X. J.
2005
Data Mining with Computational Intelligence
.
Springer
,
Berlin
.
Wenk
J.
Aeschbacher
M.
Salhi
E.
Canonica
S.
Gunten
U. V.
Sander
M.
2013
Chemical oxidation of dissolved organic matter by chlorine dioxide, chlorine, and ozone: effects on its optical and antioxidant properties
.
Environmental Science & Technology
47
(
19
),
11147
11156
.
Zhu
G.
Qin
B.
Gao
G.
2005
Direct evidence of phosphorus outbreak release from sediment to overlying water in a large shallow lake caused by strong wind wave disturbance
.
Chinese Science Bulletin
50
(
6
),
577
582
.