Journal of Environmental Protec tion, 2013, 4, 48-50
doi:10.4236/jep.2013.41b009 Published Online January 2013 (http://www.SciRP.org/journal/jep)
Study on Building and Modeling of Virtual Data of
Xiaosha River Artificial Wetland
Qun Miao, Lei Bian, Xipeng Wang, Chang Xu
School of Environment and Municipal Engineering, Qingdao Technological University, Qingdao, China.
Received 2013
ABSTRACT
From the vie wpoint of syste ms science, this artic le takes Xi aosha River artificial wetland under p lanning and co nstruc-
tion as object of study based on the systems theory and takes the accomplished and r unning pr oject of Xinxuehe artifi-
cial wetland as reference. The virtual data of quantity and quality of inflow and the quality of outflow of Xiaosha River
artificial wetland are built up according to the running experience, forecasting model and theoretical method of the ref-
erence project as well as the comparison analysis of the similarity and difference of the two example projects. The vir-
tual data are used to study the building of forecasting model of BP neural network of Xiaosha River artificial wetland.
Keywords: Xiaosha River Artificial Wetland; Virtual Data; BP Neura l N e twork
1. Introduction
To build the model of an artificial wetland, we need ab-
undant measured data. But as Xiaosha river artificial
wetla nd is und er pla nning a nd const ructio n, it’s impo ssi-
ble to obtain the experimental data. Therefore we may
use virtual experimental data to satisfy the need of mod-
eling of Xiaosha river artificial wetland. The building of
the virtual data shall be based on the similarity and dif-
ference between the object of study (Xiaosha river artifi-
cial wetland) and the reference project (Xinxuehe artifi-
cial wetland). The virtual data of Xiaosha river artificial
wetland shall be built up according to the running expe-
rience and forecasting model of the reference project and
used for building the forecasting model of BP neural
network of Xiaosha river artificial wetland.
2. Building of the Inflow Quantity Virtual
Data of Xiaosha River Artificial Wetland
According to the monitoring documents of the reference
project Xinxuehe artificial wetland, the inflow quantity
of the r u n ni ng wet la nd s ho ws an i mpo r ta n t var ia ti o n wit h
change of season and certain regularity and the real in-
flow rate floats up and down from the designed inflow
rate. Therefore, the virtual data of the inflow quantity of
Xiaosha river artificial wetland is built up virtually ac-
cording to the ratio of real inflow quantity to designed
inflow quantity and by comparing the data of the same
period (between 2008 and 2009). The virtual values of
inflow quantity are given i n Table 1.
3. Building of the Outflow Quality Virtual
Data of Xiaosha River Artificial Wetland
3.1. Modeling of Outflow Quality of the
Reference Project
It is assumed that Xiaos ha river ar tificial wetland is co m-
pletely the same as the reference proj ect. The established
forecasting model of the water quality of BP neural net-
work of the reference project Xinxuehe artificial wetland
is used to forecast the outflow quality data of Xiaosha
river artificial wetland.
Table 1. Table ty pe styles (Table capti on is indispensable).
Monitoring
month
2008 2009
Vir tu al
Inflow
quantity
（ m3/d ）
Rati o of virtua l
inflow rate/
designed
inflow rate
Virtu al in-
flow
quantity
（ m3/d ）
Rati o of virtua l
inflow rate/
designed
inflow rate
2
3
4
5
6
7
8
9
10
11
12
68200
59780
48800
55510
184830
159210
169580
117120
62830
62310
55180
2.20
0.98
0.80
0.91
3.03
2.61
2.78
1.92
1.03
2.01
1.78
76570
57340
44530
50630
123220
154330
233630
189100
175680
107570
71300
2.47
0.94
0.73
0.83
2.02
2.53
3.83
3.10
2.88
3.47
2.30
Note. : In this area, Janu ary is th e freezing pe riod and there is no inflow int o
the wetland in this month.

Study on Building and Modeling of Virtual Data of Xiaosha River Artificial Wetland
49
According to the virtual data building method, the
building of the outflow quality data of Xiaosha river ar-
tificial wetland is based on the forecasting model of the
reference project Xinxuehe artificial wetland, i.e. the
inflow data of Xiaosha river can be taken as input to
forecast the outflow quality of Xiaosha river artificial
wetland when the two projects are completely the same.
Therefore, it’s necessary to build up a forecasting model
of BP neural network of the reference project with inflow
quantity and quality as input and outflow quality as out-
put. The fore c a sting model is built up for thre e seasons of
February ~ May, June ~ September and October ~ De-
cember.
This model is in a structure of a three-layer neural
network[1], in which there’re 3 input layer neural ele-
ments and 2 output layer neural elements, the transfer
function of the hidden layer neural element is logsig, the
transfer function of the output layer neural element is
purelin, the maximum number of study cycles of the
network is 10000, the target error is 0.001, the study
speed is 0.1. For the three seasons, the number of nodes
is 9, 11 and 11 respectively and the training function of
the network is trainrp, trainscg and trainrp. The relative
error of the sample test is within the acceptable range
after testing. The relative error of the forecasting of water
quanti ty, CO DCr a nd NH3-N o f the neur al net work mod-
el of February ~ May i s 4.46% and 7.34%; the rela- tive
error of the forecasting of the model of June ~ Sep-
tember is 6.00% and 7.21%; the relative error of the fo-
recasting of the model of October ~ December is 6.94%
and 7.97% respectively.
The error of the forecasting model in each season of
the reference project is from -11.38% to 12.43%. Ac-
cording to the criterion rule o f the forecasti ng model [2]-
[3], the training result is acceptable when the relative
error is below 30%~40% during the analog forecasting of
BP neural network. Therefore, the network training is
successful and its performance can meet the requirement
of real application and can be used to build the virtual
data of Xiaosha river artificial wetland.
3.2. Building of Outflow Quality Virtual Data of
Xiaosha River Artificial Wetland
\Xiaosha river artificial wetland and the reference project
are similar and different at the same time. These differ-
ences decide that the outflow quality of the wetland is
different when the inflow quantity and quality are the
same for the two projects. Therefore, the outflow quality
virtual data o f Xiaosha river ar tificial wetla nd need to be
adjusted according to the difference between the two
projects. The adjusted data are the outflow quality virtual
data of Xiaosha river artificial wetland.
The analysis shows that the difference consists in the
optimization of the design of Xiaosha river artificial wetland,
mainly including the enlargement of wetland size, the
regularity of geometry of wetland and the reinforcement of
wetland function zone. The determination of the favor-
able impact of the three optimizing factors on the elimi-
nating rate of pollutant of the wetland in each season
needs to be obtained from the analysis of abundant ex-
perimental data. But Nansihu basin wetland ecological
recovery project is now under demonstration study phase,
so it’s only possible to make the above qualitative analy-
sis on the favorable impact of the three optimizing fac-
tor s in each se ason, but no t po ssible t o make q uanti tative
determination by using mathematics method. Therefore,
it’s necessary to transform the qualitative analysis into
quantitative analysis to realize the building of outflow
quality virtual data of Xiaosha river artificial wetland.
This article tr ansforms the qualitative ana lysis into quan-
titat ive analysi s by us ing exper t rat ing metho d.
To make the adjusting coefficient as accurate as possible,
during the study, we invited some experts who have
many years of study experience to rate for the artificial
wetland. Before rating, we provided them with the doc-
ument s re late d to the co nstr ucti on and run ning o f the two
example projects and the experts rated for the two indices
CODCr and NH3-N according to their own study expe-
rience and result [4-5].
The statistics result of the experts’ ratings shows that
the adjusting coe fficient of CODCr of t he outflo w quali ty
data of Xiaosha river artificial wetland from February to
May is 0.72; the adjusting coefficient of CODCr from
June to September is 0.42; the adjusting coefficient of
CODCr from October to December is 0.47. The adjusting
coefficient of NH3-N from February to May is 0.82; the
adjusting co efficient of NH3-N fro m June to September is
0.61; the adjusting coefficient of NH3-N from October to
December is 0.72. Therefore, the outflow quality virtual
data of Xiaosha river artificial wetland in 2008 and 2009
are gi ven in Tabl e 2.
4. Building of Neural Network Forecasting
Model of Xiaosha River Artificial Wetland
The BP neural network forecasting model apt for Xiao
sha river artif icial wetla nd is built up for three seaso ns o f
February ~ May, June ~ September and October ~ De-
cember based on the virtual data of Xiaosha river artifi-
cial wetland and using matlab software.
The forecasting model of Xiaosha river artificial wet-
land is a three-layer neural network, in which there’re 2
input layer neural elements and 3 output layer neural
elements. A fter rep etitive trial s: the nu mber of the hidd en
layer neural ele ments for model Februar y ~ May is 9; the
number of neural elements of June ~ September is 12; the
number of neural elements of October ~ December is 13.
The parameters of the forecasting model of Xiaosha river
artificial wetland are deter min ed as: t he max imum nu mber

Study on Building and Modeling of Virtual Data of Xiaosha River Artificial Wetland
50
Table 2. Outflow quality virtual data of Xiaosha river ar-
tificial wetland .
Year
Model
forecasting data
Adjusting Vir tual outflow
coefficient I data
CODCr NH3-N CODCr NH3-N CODCr NH3-N
First season (February~ May)
2008.02
2008.03
2008.04
2008.05
2009.02
2009.03
2009.04
2009.05
30.58
29.97
18.06
28.95
19.47
15.04
23.27
21.23
0.84
0.96
0.73
0.69
0.74
0.63
0.60
0.58
0.72 0.82
22.02
21.58
13.00
20.84
14.02
10.83
16.75
15.29
0.69
0.79
0.60
0.57
0.61
0.52
0.49
0.48
Second season (June~ September)
2008.06
2008.07
2008.08
2008.09
2009.06
2009.07
2009.08
2009.09
25.41
30.23
19.39
28.61
16.79
18.42
25.12
20.73
0.95
0.86
1.77
0.98
0.30
0.41
0.59
0.53
0.42 0.61
10.67
12.70
8.14
12.02
7.05
7.74
10.55
8.71
0.58
0.52
1.08
0.60
0.18
0.25
0.36
0.32
Third season (October~ December)
2008.10
2008.11
2008.12
2009.10
2009.11
2009.12
20.22
19.24
21.62
18.34
22.38
23.32
1.62
0.99
0.84
0.72
0.69
0.41
0.47 0.72
9.50
9.04
10.16
8.62
10.52
10.96
1.17
0.71
0.60
0.52
0.50
0.30
of study cycles is 10000; target error is 0.001; study
speed is 0.1; transfer function of output layer neural ele-
ments is purelin; transfer function of hidden layer ne ural
elements is logsig. The training functions for neural net-
work of each season are: trainrp for model of February ~
May, trainrp for model of June ~ September, traingda for
model of October ~ December.
The virtual data test samples are used for testing and
verification of the neural network forecasting model of
each season. The testing result shows that for Xiaosha
river artificial wetland, the relative error of water quanti-
ty of model February ~ May is 9.35%; the relative errors
of CO DCr a nd NH3-N are 12.96% and 9.55% respectively;
the relative error of water quantity of model June ~ Sep-
tember is 19.25%; the relative errors of CODCr and
NH3-N are 5.45% and 8.38% respectively; the relative
error of water quantity of model October ~ December is
12.10%; the relative errors of CODCr and NH3-N are
11.90% and 19.11% respectively, so the model training
meets t he r eq uir e me nt. B ut a ft er runnin g of X ia os ha r i ver
artificial wetla nd , it’s still necessary to get real monitore d
data to test and correct the model.
5. Conclusions
1) We ta ke the inflow quantit y and quality vir tual data
of Xiaosha river artificial wetland as input and use the
outflow quality forecasting model of the reference
project to forecast and obtain the outflow quality data o f
Xiaosha river artificial wetland when the two example
projects are completely the same.
2) According to the difference with the reference
project, we analyze the impact of the three optimizing
design factors including enlargement of wetland size,
regularity of geometry of wetland and reinforcement of
wetland function zone on t he eliminat io n rate of pollutant
and use expert rating method to make quantitative
analysis on the impact and determine the adjusting
coefficient in order to realize the building of the out flow
quality virtual data of Xiaosha river a rtificial wetland.
3) We use the virtual data test samples to test and
verify the neural network forecasting model of each
season. The test result shows that the model training
meets t he r eq uir e me nt. B ut a ft er runnin g of X ia os ha r i ver
artificial wetla nd , it’s still necessary to get real monito r e d
data to test and correct the model.
6. Acknowledgemen ts
This study is sponsored by the project of “Nansihu dete-
riorated wetland ecological recovery and water quality im-
proveement technology and demonstration” (No. 2009-
ZX07210-009) which is an important specific project of
national water body pollution control.
REFERENCES
[1] Nan Liang, “Neural Network And the Fuzzy Control,”
Beijing, Tsin ghua University Press,2000.
[2] Nan Liang, “Water Quality Forecast and Realization By
MATLAB Based On Artificial Neural Network,”
Chang’an University,2007.
[3] Xiaoping Wang, Jiyang Sun, Xin Jin, “Forecast of Water
Quality Indices of Qiantang River Based on BP Neural
Network,” Zhejiang University Journal (Engineering),
2007.41（2）, pp. 361-364.
[4] HuangXinRong, “Qian xuesen, Complexity TH oughts
study,” 2004.10,12（4）.
[5] MiaoDongSheng, “Qian xuesen Complexity Research
Review,” Journal of xi 'an jiaotong university(social
science edition), 2004.12,24（4）.