Application Methods artificial neural network(Ann) Back propagation structure for Predicting the value Of bed channel roughnes scoefficient

International Refereed Journal of Engineering and Science (IRJES)
ISSN (Online) 2319-183X, (Print) 2319-1821
Volume 4, Issue 6 (June 2015), PP.23-34
www.irjes.com 23 | Page
Application Methods artificial neural network(Ann) Back
propagation structure for Predicting the value Of bed
channel roughnes scoefficient
Wibowo1
, Suripin2
, Kodoatie2
, Isdiyana3
1
Student of Doktoral Program on University of Diponegoro,50241Semarang
1
Water Resources Engineering Department, University of Tanjungpura, 78115 Pontianak, Indonesia.
2
Civil Engineering, University of Diponegoro, 50241Semarang.
3
River Balai, Water Resources Research Center, Bali bang the Ministry of Public Works
Abstract:- Forecasting Manning roughness coefficient act out an important role in hydraulic engineering
because it is use full for the design of hydraulic structures, modeling of river hydraulic sand sediment transport.
This paper used of back propagation neural network method for predicting the Manning roughness coefficient.
Data used in the form of experimental results form the bed configuration in a laboratory and secondary data, a
total of 352data.The results using of backpropagationneural network method is optimized and accurate enough
to7-10-1network architecture, namely one input layer with 10 neurons, one hidden layer with 10 neurons and
one output layer with one neuron. Parameters used logsig activation function and function trainer of training,
with a tolerance of error of 0.01; 0.05learning rate and the maximum epochs much as1000.The model that is Q
prediksi = 0,95 Q simulasi +0,0012. With the correlation coefficientof0.980.The resulting MSE valueis0.00000177
and value for NSE of0.597.The training data as well as the value suit ability the curve of 1:1.
Keywords:- Prediction; Roughness Coefficient;Neural Network Back propagation.
I. INTRODUCTION
In engineering hydraulics, Manning roughness coefficient is an important parameter in the design of
hydraulic structures, modeling of river hydraulics and sediment transport (Bilgin & Altun, 2008; Greco et al.,
2014; Mirauda & Greco, 2014). Roughness coefficient of resistance applied to open channel flow, which is used
to calculate the velocity and flow rate (Bilgil, 2003; Bahramifar et al., 2013).
The calculating the roughness coefficient of instead be an easy task because of the complexity of the problem of
open channel. As we know that the Manning roughness coefficient roughness coefficient representing the
resistance of flow by applying the flow in the channel. There for eroughness coefficientis also a fundamental
parameter of fluid flow calculations that is still highly demanded in its application (Bilgil &Altun, 2008).
Resistance of flow in alluvial channels with relatively high accuracy is also a concern for the hydraulic
engineer. However, the problem is still unsolved despite numerous investigations over the last few decades
(Yang &Tan, 2008). Among the problems are due to changes in channel form the bed configuration, the aspect
ratio of the depth and width, the influence of the side wall, and the wall shear stress sisnotuni formally
distributed in the three-dimensional shapes due to the presence of the free surface and the secondary current
(Azamathulla et al., 2013; Samandar, 2011; Bilgin & Altun, 2008; Yang &Tan, 2008; Guo & Julien, 2005).
Along with the growing world of digital (computer), some of models have been developed to simulate this
process. Neither the empirical model (black box model), conceptual model (physical process based), the model
continuously (continuous events), lumped models, distribution models and models of single (Setiawan &
Rudiyan to, 2004). These models are formed by a set of mathematical equations that reflect the behavior of the
hydro logical parameters, so the parameters contained in the equation has a physical meaning (Adidarma, et al.,
2004).
The last few years, art official neural networks (ANN) as a form of black box model (black box model),
has been successfully used optimally to model non-linear of input-output relationship in a complex hydro logic
processes and the potential to become one of the decision-making tool promising in hydrology (Dawson and
Wilby, 2001). ANN is a form of artificial intelligence that has the ability to learn from the data and does not
require a long time in the making models (Setiawan & Rudiyan to, 2004).
These models uses mathematical equations of linear and non-linear that do not take into account at all physical
processes, but the most important in this model is the output produced by the actual approach (Adidarma, et al.,
2004). In addition, the ANN was also able to identify the structure and also effective in connecting the input and
output of simulation and forecasting models (Setiawan and Rudiyan to, 2004).

Application Methods artificial neural network (Ann) Back propagation structure for Predicting the value
The ability of Artificial Neural Networks (ANN) in solving complex problems has been demons trated
in various studies, recently, the development of the body in the application of artificial neural net working river
engineering like Karunanithi et al. (1994), Fauzy & Trilita (2005), Cigizoglu (2005), Antenatal. (2006), Bilgil&
Altun (2008), Samandar (2011), Mary (2011), The control of the water level (Alifia et al., 2012), Azamathulla et
al. (2013), and Bahramifar et al. (2013), Model is as irainfallrun off (Doddy & Ardana, 2013), rainfall prediction
in Jakarta(Nugroho et al, 2013).
Therefore, this paper will apply the method of ANN. The purpose of this paper is to use the approach
of artificial neural network (ANN) for calculating the Manning roughness coefficient using data from laboratory
experiments. In the study raised the flow parameter measurements in the laboratory is used for artificial neural
network as input parameters. The value of calculating the roughness coefficient is calculated later Maning used
to estimate the flow in open channel flow.
II. MATERIALS AND METHODS
2.1 Coefficient of Roughness in Open Channels.
Ina lot of literature it is known that velocity offlowin open channels formulation created by the Robert
Manning (1891), as Equation (1)
V =
1
n
R2/3
S .....................................(1)
Where Vis the average velocity of the cross section, Nis Manning resistance co efficient, Ris the hydraulic
radius and Sis the hydraulic slope. This formula is derived from semi-empirical that has been used hydraulic
experts during the 18th century.
Dischargeormagnitude of the flow of the river/canal is flowing through the volume flow through ariver cross
section/channel per unit time (Chow, 1959; Soewar no, 1995). Usually expressed in units of cubic meters per
second (m3 / s) orlitersper second (l /sec). Flowis the movement of water in the river channel/channels.
Basically discharge measure mentis a measurement of wet cross-sectional area, the flow rate and water level.
The general formula is used as Equation (2).
Q = V. A ..........................................(2)
Today, Manning equation is more often used as a formulation in hydraulic engineering and expressed respective
lyin Equation (3).
n =
1
V
R2/3
S .................................(3)
Formulation development at Manning formula also applied to the linear separation method. This linear
separation method has been widely recognized by experts as a hydraulic principles and approaches on the sum
of components resistance. Resistance to the flow in the channel digolongan into 2 (two) types, the first friction
surface (skin friction) that is generated by the boundary surface resistance and depending on the depth of the
flow relative to the size of the elements on the surface roughness limit, both opposition form (form resistance) or
form drag namely roughness related to the geometry of the surface roughness of granules and barrier forms
associated with the basic configuration that govern vortex and secondary circulation. This principle has been
developed in a natural resistance component with rigid base and a natural resistance component with a flexible
base (Meyer-Peter & Muller, 1948; Einstein and Barbarossa, 1952; England, 1966; Smith & McLean, 1977;
Griffiths, 1989; Yang & Tan 2008).
Manning equation formulation in linear separation method as Equation (4).
𝑛 = 𝑛 𝑤 + 𝑛′
+ 𝑛′′
……………….(4)
Where𝑛 𝑤 isroughness coefficientdue tothe sidewall, with𝑛 𝑤 =
𝑅1/6
𝑔
𝑢∗𝑤
𝑈
and𝑢∗𝑤 = 𝜏 𝑤 /𝜌. 𝑛′
is the resistance
due to friction surfaces(skin friction) or the roughness of granules, the formulan
𝑛′
𝑅1/6 𝑔 =
𝜏0
′
𝜌
𝑈
=
𝑢∗
′
𝑈
and𝑛′
′ is
the resistance that is due to form drag(form drag) orroughnessshape, with the formulation ofn with
𝑛′′
𝑅1/6 𝑔 =
𝑢∗
′′
𝑈
. Equation (3) on a restated as a function of dimension lesssymbolonan open channel roughness
coefficient(
𝑛′′
𝑛′ )as in Equation(5).
𝑛′′
𝑛′ = 𝑓 (𝑅 𝑒, 𝐾𝑟 , 𝜂, 𝑆, 𝐹𝑟, 𝜎,
𝑏
𝑕
) ................(5)
Where 𝑅 𝑒is the Reynolds number, 𝐾𝑟 is relative roughness usually expressed as𝑘 𝑠/
𝑅 where𝑘 𝑠isequivalentwallsurface roughness, ηis across-sectional geometries, S is channels lope, 𝐹𝑟is the
Froude number and σ is the gradation grain. In the Equation (5) are further tested from the description of the
mechanism and limit the flow channel by Yen (2002&1992). Symbol function in Equation (5) is not linear and

complex. For the sake of simplification made in the conventional approach. As noted, the problem of flow in
open channels may be completed with an error limit of ± 10% (Bilgil, 1998).
These indications show the new and accurate methods are still needed. The existence of the methods that have
high accuracy will reduce error rates. At the ends of the artificial neural network approach to the efficiency of
the pre-assessment approach to predict the roughness coefficient through the use of Artificial Neural Network
(ANN).
2.2 Artificial Neural Network (ANN).
Neural Network (NN) is a learning method that is inspired by the biological network learning system
that occurs on the network of nerve cells (neurons) are connected with one another. NN structure used is Back
propagation (BP) which is a systematic method for training multiplayer. This method has a powerful
mathematical basis, objective and this algorithm to get the form of equation and the coefficient in the formula by
minimizing the number of error squared error by the model developed in the training set (Bilgil & Altun, 2008).
2.3 Algorithm a Back propagation (BP)
Back propagation algorithm on neural network (BPPN) is a systematic approach to training
(calibration) on multi layer percept on neural networks or multilayer (multilayer perceptrons). Layer (layer) The
first consists of a set of inputs and the final layer is the output (target). Among the input layer and output layer
there is a layer in the middle, which is also known as hidden layers (hidden layers), could be one, two, three and
so on. In practice, the number of hidden layers is at most three layers. Input layer mere present asking input
variables, hidden layer represents non linearity(non-linearity) of the network system while the output layer
contains variable output, the last layer output from the hidden layer directly used as the output of the neural
network.
BP training process requires three stages of data input for training feed forward, back propagation to the value of
the error (the error) as well as the adjustment of the weight values of each node of each layer on ANN.
Beginning with feed forward value input, each input to the unit-i(xi) receiving an input signal which will then be
transmitted to the hidden layerz1, ..., zp. Furthermore, the j-the hidden unit will calculate the value of the signal
(zj), which will be tram smitted to the output layer, using the activation function (f)
In simple terms of BPNN described as follows, an input pattern in corporate into the network system to produce
output, which is the compared with the actual output pattern. If there is no difference between the output of the
system and the actual network, then the learning is not necessary. In other words, a weight that indicates the
contribution of input node to hidden nodes, as well as from hidden node to output, in which case the difference
(error) between the output of the system with the actual network, then the weights repaired one backwards, from
the output passes the rough hidden node andr e-input node. Mathematically can be described in the back
propagation algorithm in Equation(6).
𝑧𝑖𝑛 𝑗
= 𝑣0𝑗 + 𝑥𝑖 𝑣𝑖𝑗
𝑛
𝑖=1 ...........................(6)
where 𝑧𝑖𝑛 𝑗
is aktifivasifunctionto calculate thevalue ofthe outputsignalin thehidden nodej; 𝑥𝑖isthe valueinthe
inputnode ; 𝑣𝑖𝑗 = is theweight valuethat connects theinputnode iwith ahiddennotej. ; 𝑣0𝑗 is
avaluebiaswhichconnects thebiasnode1with thehidden nodej.nisthe number of inputnodesin the inputlayer.
And the output signal from the hidden node j given sigmoid activation function as Equation(7)
𝑧𝑖 = 𝑓(𝑧𝑖𝑛 𝑗
) =
1
1+𝑒
− 𝑧 𝑖𝑛 𝑗
...........................(7)
where𝑧𝑖isthe signaloutputofhidden nodej. While each unit of output k(Yin),as Equation(8)
𝑌− 𝑖𝑛 = 𝑤0𝑗 + 𝑧𝑗 𝑤𝑗𝑘
𝑝
𝑗=1 ...........................(8)
And the activation function to calculate the value of the output signal, as Equation(9)
𝑌 = 𝑓(𝑌− 𝑖𝑛) =
1
1+𝑒−𝑌− 𝑖𝑛 .............................(9)
During the training process progresses, each unit of output compares the target value. for a given input pattern
for calculating the value of the parameters that would improve(update) the weight of the value of each unit in
each layer(Hertz et al., 1991). Nodes in the output layer have a valuebetween0-1.
2.4 Artificial Neural Network (ANN) in Determining a Bed Roughness Coefficient.
In this paper, the calculation of roughness coefficient in open channels is performed using Multilayered
Perception (MLP) artificial neural network. In the literature more likely to use the MLP learning algorithms to
back propagation algorithm Rumelhart et al. (1986). In this algorithm optimization of weights during the
learning process that can use the latest formulation weight given as the output function (level of movement) of
the brain (neurons).

2.5 Performance Model.
Performance models used to measure the accuracy of the model. In this paper, the performance of the model is
used to determine the degree of correspondence between the actual data with the results of forecasting used
measure of correlation coefficient, with the formula in Equation (10).
𝑅 =
𝑥𝑦
𝑥 𝑦
....................................(10)
Where= 𝑋 − 𝑋,𝑋 isthe actualdischarge,𝑋is the averagevalue ofX, 𝑦 = 𝑌 − 𝑌, Y is a discharge or as
imulation result of forecasting, 𝑌, is the average value of the Y value of correlation can be seen in Table.1
Table1. Correlation Coefficient Values
Correlation Coefficient(R2
) Iimplication
1 perfect positive
0,6 <R2
≤1 Good positive direct
0 <R2
≤0,6 Direct weakly positive
0 There is norrelationship
-0,6 ≤R2
< 0 Weak negative direct
-1 ≤R2
< -0,6 Negativestraightgood
-1 negative perfect
Source: Soewarno, 1995.
The media n square error (mean square error, MSE). MSE is a measure of the accuracy of the model by
squaring the error for each point of data in a data set and the no btain the average or median value of the sum of
the squares. The formulation of MSE as Equation(11)
𝑀𝑆𝐸 =
𝑦 𝑖− 𝑦 𝑖
𝑁
𝑖=1
2
𝑁
=
𝑒𝑖
2𝑁
𝑖=1
𝑁
...............................(11)
Where 𝑦𝑖isthe actual valueof data, (𝑦𝑖) is the value o f the results of forecasting, Nis the number of data
observations, and𝑒𝑖is per-point error data. Then used a common procedure error calculating per-point data,
which for the time series followed formulation is: data = pattern + errors for easy, error(error) is written with an
e, the data with the data pattern of X and X. In addition, the sub script i (i =1,2,3, ..., n) are included to show the
data point to-i, so written𝑒𝑖 = 𝑋𝑖 − 𝑋. If youjustwant to know the magnitude of the error regardless of the
direction it is called absolute error or𝑒𝑖 = 𝑋𝑖 − 𝑋
Another criterion is the accuracy of the model or Nash Sutcliffe Model Efficiency Coefficient (NSE). Nash
gives a good indication for matching of 1:1between simulations and observations. Formulation of Nash as
Equation(12).
𝑁𝑆𝐸 = 1 −
𝑄 𝑜𝑏𝑠 −𝑄 𝑠𝑖𝑚
2
𝑄 𝑜𝑏𝑠 −𝑄 𝑜𝑏𝑠
2 .....................................(12)
WhereQobs areobservational data, Qobs is the averageobservational dataand𝑄𝑠𝑖𝑚 is the value ofthe simulation
results. NSE value criteria can be seen in Table(2).
Table 2 Criteria Value Efficiency Model Nash Sutcliffe Coefficient (NSE).
Nash Sutcliffe Model Efficiency
Coefficient (NSE) Value
Interpretasi
NSE > 0,75 Good
0,36 <NSE  0,75 Ssatisfy
NSE  0,36 Not satisfactory
Source : Motovilov et al., 1999
2.6 Data from Experiment
This paper aims to analyze the performance of artificial neural network back propagation method in
predicting the bottom friction coefficient. Writer wanted to know how the performance of artificial neural
networks back propagation method to recognize patterns of data parameters that‟s lope, depth, grain and flow.
The data used for learning and then to evaluate the use of ANN obtained experiment. The data will be used by
the laboratory results of several researchers and the results of its own research, the data include:
1. DataexperimentalWangandWhite(1993).
2. Data fromexperimentsGuyetal. (1966).
3. Research data fromSisingih(2000).
4. The result ofthe experimentfrom Wibowo(2015)

Table3. Results of Research Data
Parameter Guyetal. (1966). Wibowo(2015) Sisingih(2000) Wang
White(1993).
Slope(S) 0,00015-0,0101 0,006-0,0100 0,007-0,013 0,00001-0,00305
Discharge (Q) (m3
/s) 0,028 – 0,643 0,003-0,008 0,003-0,006 0,024-0,410
Ratio (b/h) 2,247-42,105 3,587-9,524 0,667-1,000 3,288-19,335
Velocity(V)
m/s
0,212-1,898 0,132-0,411 0,214-0,429 0,105-1,318
Reynolds Numbers (Re) 2,157-98,753 14,446-50,29 0,003-29,211 4,35-11,42
Froude Numbers (Fr) 0,089-1,714 0,152-0,324 0,194-0,353 0,073-1,049
Fricative
()
0,0015-1,734 0,291-0,842 0,727-1,982 0,021-4,685
Roughness coefficient
(n)
0,010-0,040 0,011-0,026 0,012-0,042 0,015-0,028
Sample 269 40 16 64
2.7Mapping Neural Network in the Roughness Coefficient.
In a study of open channel flow roughness coefficient and the relationship between flow parameters
will be given to the function as Equation (5).
𝑛′′
𝑛′ = 𝑓 (𝑅 𝑒,
𝑑𝑠
𝑕
, 𝑆, 𝐹𝑟, 𝜎,
𝑏
𝑕
, 𝜏∗)
Where 𝑅 𝑒 is Reynolds number = 𝑢∗ 𝑑 𝑠/𝜈 , ν is the kinematic viscosity, 𝑑 𝑠is granular particles (mm), h is the
average depth, S is the slope of the elongated base channel, 𝐹𝑟 is Frounde number 𝑉𝑟 𝑔𝑕 , 𝑉𝑟 = average speed
of the flow (Q/A), Q is the flow rate, σ is the gradation grain, with𝜎 =
1
2
𝑑84
𝑑50
+
𝑑50
𝑑16
b is the channel width (m),
𝜏 is the shear stress = ghS, S slope hydraulic, g is the acceleration of gravity and R is radius hydraulic. With
the data in Table (3.1) as a measurement parameter input (input) and output (output) is written in pairs on a set
of data created. This data set is used to calculate the roughness coefficients using Manning formula. For learning
in artificial neural network, the parameters on the right side of the symbol Equation (2.5) is given as input and
roughness coefficient as a target parameter. In the learning process half of the data set used for artificial neural
network learning, the time remaining is used to evaluate the implementation of the artificial neural network
learning.
Input data consists of relative roughness(𝑋1), Reynolds number(𝑋2), Slope(𝑋3),the Froude number(𝑋4),
gradation grain (𝑋5), the depth-width ratio(𝑋6)andshear stress(𝑋7).
2.8Network Architecture in Neural Network
In the analysis of roughness coefficient prediction of basic channels, network architecture used is many
layered network architecture (multilayer) as in Figure(1).
Figure1.Network Architecture with Back propagation

Specification: X is input nodes in the input layer; Z is hidden node (hidden layer); Y is the output node in the
output layer; 𝑉1,1,…,𝑉𝑛 isthe weightofthe inputlayerto thehidden; 𝑊1,1,…,𝑊𝑛 is the weight of the hidden layer to
the output ; 𝑏 𝑣isbiasfromthe inputlayerto thehiddenlayer; 𝑏 𝑤 isbiasof thehiddenlayertothe output layer.
2.9Training Process
The training process was conducted on the data as input parameters of network nodes ;
Toleransi error = 0,01; Learning Rate (α) = 0,5; number of iterations = 1000 times
III. RESULTS AND DISCUSSIONS.
3.1 Experiment Results
Determination parameters of the neural network is done by searching for the best value of the hidden neurons
are used. Furthermore, to facilitate the calculation of the iteration process and running experiment data, then use
the software MATLAB. Here are the results of the experiments have been conducted to determine the number of
neurons in the hidden layer Table(4).
Table 4. Comparison of Results of Experiments on Bed Relative Value Roughness Coefficient (n '/ n')
Running Arsitektur
Jaringan
Function
Activation
MSE Correlation
Coefficient
1 7-10-1 logsig 0,0102 0,908
2 7-9-1 logsig 0,0188 0,928
3 7-8-1 logsig 0,0292 0,920
4 7-7-1 logsig 0,0295 0,911
5 7-6-1 logsig 0,0566 0,909
6 7-5-1 logsig 0,0202 0,915
7 7-4-1 logsig 0,0360 0,916
Based on several the experiments that have been done, the architecture of artificial neural networks for the
prediction of roughness coefficientter optimal basic channel sare7-10-1architecture. This architecture consists of
one input layer with10neurons, onehiddenlayerwith10neuronsandoneoutput layer with one neuron. The resulting
MSE value is 0.0102andthe correlation coefficient 0.908. Pictures of the jar in gent architecture can be seen in
the Figure (2) and (3).
Figure 2.Architecture Data Network with Back propagation

Figure 3. Neural Network Training with Matlab
3.2 Development of Model
The next process is to develop a model that is based on the data Equation (2.5). by using multiple linear
regression method. Results of the model development using basic rough ness relative (n "/n ') in de pthandr
elativegraincan be seen in Equation (13) and(14).
𝑛′′
𝑛′ = 0,378 + 64,621𝑅 𝑒 − 0,537
𝑑𝑠
𝑕
− 0,00242 𝑆 − 0,005𝐹𝑟 − 0,065𝜎 + 0,008
𝑏
𝑕
+ 0,152 𝜏∗ ..(13)
where the correlation coefficient(R2
) of 0,602.MeanSquareError(MSE) of
4,565andValueEfficiencyModelNashSutcliffeCoefficient (NSE) of 0,597.
𝑛′′
𝑛′ = 0,167 + 64,876𝑅 𝑒 − 0,549
𝑑𝑠
𝑕
− 0,00013 𝑆 − 0,0037𝐹𝑟 − 0,0284𝜎 + 0,0034
𝑏
𝑕
+ 0,136 𝜏∗...(14.)
The correlation coefficient(R2
) of 0.634, Mean Square Error(MSE) of
4.805andValueEfficiencyModelNashSutcliffeCoefficient (NSE) of 0.625.
The formulation of the the oreticallinearseparation(n "/n ') can be seena sEquation(15)
𝑛′′
𝑛′ = 0,1167 𝑛 + 0,0182 ...................................(15)
By giving the value of the correlation efficient(R2
) 0,020 Mean Square Error(MSE) of
0.000254andValueEfficiencyModelNashSutcliffeCoefficient (NSE) of 0.02.
The formulation on the basis of Manning rough ness(n) can be seen as Equation(16)
𝑛 = 0,0287 + 0,104𝑅 𝑒 − 0,013
𝑑𝑠
𝑕
− 1,0187 𝑆 + 0,002𝐹𝑟 − 0,0017𝜎 + 0,0001
𝑏
𝑕
− 0,0015 𝜏∗...(16)
Generate the correlation coefficient (R2
) of 0.411, Mean Square Error (MSE) of 0.000757andValue Efficiency
Model Nash Sutcliffe Coefficient (NSE) of 0,955.By using the default batch algorithm a back propagation,
iteratingthecalibrationstage (training) until the model is designed to stop, the model usedepoch1000 times with a
time of 2 (two) seconds for each model. Calibration is a process oradjustingthe values of the parameters of
model to achiev eabestmatchbetweenobservationsand variables predictive variables. Match betweenthe actual
data(observations) withforecasting resultsbased onthe results ofthe calibrationis indicated bythe correlation
coefficient. The values ofthe weightsof thenetworkANNachievedfromthe calibrationresultswill be usedin
theverification phase(testing). the accuracy offorecasting results, performancecriteriarepresentedbythe model

Figure 4.Simulation Results between data Flume and Simulation Discharge

Simulation models according Refsgaard (2000) is an effort to valid at the use of the model to gain knowledge or
in sigh to far reality and too btain forecasts that can be used by waterre source managers. Simulation stage is the
final process after the process of calibration and verification carried out. In this stage, the overall roughness
coefficient data is used as input data to calculate the flow rate. By using the method of linear separation then
compared with the results of the discharge flume as in Figure(5) andproducea modelin Equation(17).
𝑄 𝑓𝑙𝑢𝑚𝑒 = 0,436 𝑄0,8834
𝑆𝑒𝑝𝑎𝑟𝑎𝑡𝑖𝑜𝑛 𝑙𝑖𝑛𝑒𝑎𝑟
.............(17)
Figure 5 The Relationship between the Discharge data and Flume Simulation
The results of simulations of discharge and discharge flume, asin Table (5).
Table 5 Correlation between Discharge Simulation Results and debilitated Flume.
Model Development Correlation Coefficient
(R2
)
MSE NSE
ANN method 0,980 1,7 E-06 0,844
Linear separation
method
0,859 4,324 0,774
Analysis of the
relative dimensions
of grain
0,378 4,817 0,287
Analysis of the
relative depth
dimension
0,377 4,817 0,388
Source of data: the results of calculations.
3.3 Discussion.
•The Pre Data Analysis.
Data roughness coefficient of line arseparation results (n) cannot be used directly but through the calculation of
flow rate. In connection with the separation of each line a roughness coefficient values should be described in
advance to obtain the value n 'and n"as well as due to the influence of the side wall shear. The influence of the
shape of the base konfiguasri very important, so the need for separation of the regime and form the basis of the
conditions that occurred.
•Input Data Model Analysis on Bed Roughness Coefficient Value
Input data consists of the relative roughness(X1), Reynolds number(X2), Slope(X3), the Froude number(X4),
gradation of grain (X5), the depth-width ratio(X6) and shear stress(X7) on the modeling coefficient s roughness
obtained the following results
𝑛′′
𝑛′ = 0,378 + 64,621𝑅 𝑒 − 0,537
𝑑𝑠
𝑕
− 0,00242 𝑆 − 0,005𝐹𝑟 − 0,065𝜎 + 0,008
𝑏
𝑕
+ 0,152 𝜏∗ ............(18)

With a correlation coefficient (R2
) of 0.602(0.6 <R2
<1) which shows the relationship between input variables
(independent variables) have a positive direct relationship Good. This means that data is correlated. Mean
Square Error (MSE) amounted to4.565% (<5%). Error below 5% indicates that the error between the actual
model and simulation is below to lerance. ValueEfficiencyModelNashSutcliffeCoefficient (NSE) of 0,597(0,36
<NSE 0,75) which shows that the interpretation between actual and simulated models in satisfactory
condition, or can be correlated with Good. Similarly, also by using the relative depth. Thus, the data input can
be used on the production model of the channel bottom friction coefficient.
•Separation of Linear Model Analysis on the Bed Roughness Coefficient Value.
The model obtained in the the oreticallinearseparation method(n "/n ') can be seenas Equation(19).
𝑛′′
𝑛′ = 0,1167 𝑛 + 0,0182 ...................(19)
By giving the value of the correlation coefficient(R2
) 0,020(0 <R2
<0.6) which shows the relationship
between input variables (independent variables) have a direct relationship weakly positive. This means that data
is correlated poorly. Mean Square Error (MSE) of 0,000254(<5%). Error below 5% indicates that the error
between the actual model and simulation is below to lerance and very Good.
ValueEfficiencyModelNashSutcliffeCoefficient (NSE) of 0,02(NSE <0.36) which shows that the interpretation
of the actual model and simulation in less than satisfactory condition or less correlated is good.
• Analysis on the Model Manning the Bed Roughness Coefficient Value
The model formulation Manning invitation dimensional analysis(n) can be seen as Equation(20)
𝑛 = 0,0287 + 0,104𝑅 𝑒 − 0,013
𝑑𝑠
𝑕
− 1,0187 𝑆 + 0,002𝐹𝑟 − 0,0017𝜎 + 0,0001
𝑏
𝑕
− 0,0015 𝜏∗ ...............(20)
By giving the value of the correlation coefficient (R2
) 0.411(0 <R2
<0.6) which shows the relationship between
input variables (independent variables) have a direct relationship weakly positive. This means that data is
correlated poorly. Mean Square Error (MSE) of 0,000757(<5%). Error below 5% indicates that the error
between the actual model and simulation is below to lerance and very good. Value Efficiency Model Nash
Sutcliffe Coefficient (NSE) of 0,955(NSE>0.75) which shows that the interpretation of the actual model and the
simulation under condition scorrelate well.
•Analysis on Manning on Flow Model.
Based on the Table(4.2) obtained results for the model artificial neural network(ANN) have satisfactory
results, whether of the correlation between the variablesof0,980(0,6 <R2
<1) which shows the relationship
between input variables (independent variables) have a relationship strong positive immediately. This means
that the data correlates very well. Mean Square Error (MSE) of 0,00000177(<5%).
Error below 5% indicates that the error between the actual model and simulation is below to lerance
and shown very good relationship. For the best fore casting method is the method that produces the smallest
error. Value Efficiency Model Nash Sutcliffe Coefficient (NSE) of 0.597(0.36 <NSE 0.75) which shows that
the interpretation between actual and simulated models in satisfactory condition, or can be correlated with
either. Similarly, also by using the relative depth.
Similarly, the flow separation method that shows the results of the correlation between variables in the
model𝑄 𝑓𝑙𝑢𝑚𝑒 = 0,436 𝑄𝑠𝑖𝑚𝑢𝑙𝑎𝑠𝑖
0,8834
with R2
=0,859 (0,6 <R2
< 1) which shows the relationship between input
variables (independent variables) has a direct relationship strong positive. This means that the data correlates
very well. Mean Square Error (MSE) of 0,00000177(<5%). Error below 5% indicates that the error between the
actual model and simulation is below to lerance and devoted relationship very well. For the best fore casting
method is the method that produces the smallest error. Value efficiency model Nash Sutcliffe coefficient (NSE)
of0,774(NSE >0,75) which shows that the interpretation of the actual model and the simulation under condition
scorrelate well.
Whereas the method of analysis dimensions that are less good result.
IV. CONCLUSION
Based on the analysis and discussion it can be concluded as follows:
1. Utilization of ANN to the practical application of basic channels for ecastin groughness coefficient
generally reliable.
2. The best results from the model ANN depends on the quality of the data, including in this case the length
of the data so that the model ANN is able to perform pattern recognition in put and output relationship.
3. ANN Back propagationbestregression modelbasedinputwitharchitecture7- 10 -of 1 (7 units of inputto
theinputlayer-10hidden unitsin the hidden layer-1 unitof outputin the output layer) is moreaccuratelyused
inforecastingkoefsienbasicroughnessandprovedmore capablefollowthe characteristicsofthe actual

datawiththe value ofthe correlationbetween variablesat0.980 with modelobtainedthe𝑄 𝑝𝑟𝑒𝑑𝑖𝑘𝑠𝑖 =
0,95 𝑄𝑠𝑖𝑚𝑢𝑙𝑎𝑠𝑖 + 0,0012,NSEvaluesof 0.844andMSE of0.00000177.
4. Modellinearseparationcan beused to estimate theflow rateto the conditionsforthe basic shapeof channels.
5. Linearseparation modelcanbe usedto estimatethe flow rate bythe basicconditions oftheirshapeto the shape
ofthe channel𝑄 𝑓𝑙𝑢𝑚𝑒 = 0,436 𝑄𝑠𝑖𝑚𝑢𝑙𝑎𝑠𝑖
0,8834
with (R2
=0.859).
ACKNOWLEDGMENTS
Experimental work carried out in the Central Solo River, Indonesia. The author would like to thank the
Balai of Solo River in Central Java province, which has been providing information and data-the data to support
this research and The author would like to acknowledge the assistance of Surip in, Robert Kodoatie, Isdiyana,
Kirno and Family in conductingex periments too. Special thanks to Uray Nurhayati, Ajeng, Hanif and Amira for
their help during the work.
REFERENCES
[1]. Adidarma, W.K., Hadihardaja, I.K., and Legowo, S. (2004). “Rain-Runoff Modelling Comparison
Between Artificial Neural Network(ANN) andNRECA". ITBCivil Engineering Journal, Vol. 11No.3:
105-115
[2]. Alifia, F.A, Triwiyatno, A., and Wahyudi., 2012. “Control System Design of Adaptive Neuro Fuzzy
Inference System (ANFIS) “Case Study: Water Front Pengontoran Altitudeand Temperature Steam
Boiler Steam Drum, transient Journal, Vol.1, No.4, 2012, ISSN: 2302-9927, 312, Semarang.
[3]. Altun H, Bilgil A, Fidan CF. Treatment of multi-dimensional data to enhance neural network
estimators in regression problems. Expert Systems with Applications 2006;32(2):Available online from
http://www.sciencedirect.com/.
[4]. Azamathulla, H. Md., Ahmad Z., and Aminuddin Ab. Ghani, 2013.” An Expert System for Predicting
Manning‟s Roughness Coefficient in Open Channels by Using Gene expression Programming “ Neural
Comput & Applic, 23:1343–1349.
[5]. Bahramifar, A, Shirkhanib,R., and Moham madic, M., 2013. ” An ANFIS-based Approach for
Predicting the Manning Roughness Coefficient in Alluvial Channels at the Bank-full Stage “,IJE
Transactions B: Applications Vol. 26, No. 2, pp.177-186.
[6]. Bilgil, A and Altun, H., 2008. “ Investigation of Flow Resistance in Smooth Open Channels using
Artiﬁcial Neural Networks”. Flow Meas Instrum 19:404–408.
[7]. Bilgil, A., 2003. “Effect of Wall Shear Stress Distribution on Manning Coefficient of Smooth Open
Rectangular Channel Flows “,Turkish J. Eng. Env. Sci.27 , pp. 305 – 313.
[8]. BilgilA., 1998.” The Effect of Wall Shear Stress on Friction Factor in Smooth Open Channel Flows,
Ph.D.Thesis.Trabzon (Turkey):Karadeniz Technical University,Turkish.
[9]. Chow, V.T., 1959,” Open Chanel Hydraulics”, Mc Graw Hill Kogakusha, Ltd.
[10]. Cigizoglu HK. “ Application of Generalized Regression Neural Networks to Intermittent Flow
Forecasting and Estimation. Journal of Hydrologic Engineering, 2005,pp.336–341.
[11]. Dawson, C.W. and Wilby, R.L. (2001). “Hydrological Modelling Using Artificial Neural Networks”.
Progress in Physical Geography, 25-1: 80-108.
[12]. Doddy. P., and Ardana. Heka., 2013. “Artificial Network Application Requirements (arificial Neural
Network) In Model isasi Rainfall Runoff with Comparing theTraining Algorithm(Case Study:
DASTukadJOGADING, 139A), the National Conference of Civil Engineering7(context 7), pp.A107-
A114. Elevenuniversitiesin March(UNS) –Surakarta
[13]. Einstein, H A., and Barbarossa, 1952. “ River Channels Rougness .” Trans. ASCE, 117, pp.1121-
1132.
[14]. Engelund, F., 1966." Hydraulic Resistance of Alluvial Streams." J.Hydr.Div.,ASCE, 92 (HY2), 315-
326.
[15]. Fauzi, M and Trilita,M,N., 2005.” Application sarificial Neural Network for Flow Forecasting Aungai
Blega". Jurnal Engineering Planning, Vol1, No.3.UPN Veteran, Jawa Timur.
[16]. Greco, M., Mirauda., D., and Plantamura, V. A., 2014.” Manning‟s Roughness Through the Entropy
Parameter for Steady Open Channel Flows In Low Submergence” 2th International Conference on
Computing and Control for the Water Industry, CCWI2013, ScienceDirect. Procedia Engineering 70
.pp773 –780.
[17]. Griffiths., G. A., 1989. “ Form Resistance In Gravel Channels With Mobile beds “.Journal of
Hydraulic Engineering, Vol. 115, No. 3,© ASCE. Paper No. 23245.pp 340-355.
[18]. Guo, J. and Julien, P.Y., 2005. “Shear Stress in Smooth Rectangular Open-Channel Flows”, Journal of
Hydraulic Engineering,ASCE, Vol. 131 , No. 1,pp. 30–37.

[19]. Hertz, J.; Krogh, A.; Palmer, R. G., 1991. “ Introduction to the Theory of Neural Computation,
Redwood City, CA: Addison Wesley.
[20]. Karunanithi N, Grenney WJ, Whitly D, Bovee K. Noural networks for river flow prediction. Journal of
Computing in Civil Engineering 1994;8. Pp. 201–220.
[21]. Meyer, Peter, and Muller, 1948. “ Formula for Bed-Load Transport ”, Ins.Ass. Hydr.Res. 2nd
Meeting,
Stockholm
[22]. Mirauda, D., Greco, M., and Moscarelli, P., 2011. “ Entropy Based Expeditive Methodology for Rating
Curves Assessment”. Proceedings of the International Conference on Water, Energy and Environment,
Phuket, Thailand, 1351-1356.
[23]. Motovilov, Y.G., Gottschalk, L., Engeland, K. & Rodhe, A. 1999. Validation of a Distributed
Hydrological Model Against Spatial Observations. Elsevier Agricultural and Forest Meteorology. 98,
pp. 257-277.
[24]. Nugroho, J.T., Liong,T.H., Safwan Hadi and Bayong Tjasyono HK., 2013.”
IncreasedMonthlyRainfallPredictionAccuracyinJakartaRegionalDataUsingTropicalRainfallMeasuring
Mission(TRMM) andNetwork-Based DataCosmic RaysTermsArtificial, Journalof Aerospace Sciences,
Vol. 11No.1, pp41-48,Jakarta.
[25]. Refsgaard, J.C. 2000. Towards a Formal Approach to Calibration and Validation of Models Using
Spatial Data, Dalam R. Grayson & G. Blöschl. Spatial Patterns in Catchment Hydrology:
Observations and Modelling. Cambridge University Press, Cambridge, pp. 329 – 354.
[26]. Rumelhart DE, Hinton GE, Williams RJ, 1986. “ Learning internal representation by error propagation.
In: Parallel and distributed processing: Explorations in the microstructure of cognition. Vol 1:
Foundations. Cambridge, MA: MIT Press.
[27]. Samandar. A., 2011.” A Model of Adaptive Neural-Based Fuzzy InferenceSystem (ANFIS) for
Prediction of Friction Coefficient inOpen Channel Flow “, Scientific Research and Essays Vol. 6(5),
pp. 1020-1027.
[28]. Setiawan, B.I., and Rudiyanto. (2004). “ Applicationof NeuralNetworkstoPrediction ofWatershed",
Proceedings ofWorkshopSimulationandComputingTechnologyandApplications2004 -BPPT, Jakarta.
[29]. Soewarno, 1995. “HydrologyApplicationsStatistical MethodsforDataAnalysis, Volume1and2",
PublisherNovaBandung.
[30]. Sisingih, D., 2000, “Relationship Study of Bedform on Hydraulic roughness coefficient on the Open
Channel ", Thesis Master of Civil Engineering, ITB, Bandung.
[31]. Smith, J.D. and McLean, S.R., 1977. “Spatially Averaged Flow Over a Wavy Surface ”. J. Geophys.
Res., 84(12), 1735-1746.
[32]. Wibowo, H., 2015. “ Manning Roughness Coefficient Study on Bed Materials Non-Cohesive with
Parameters Using Entropy to Open Channel Flow”, Proceedings of International Conference : Issues,
Management And Engineering In The Sustainable Development On Delta Areas Semarang, Indonesia
– February 20th, 2015Paper No. XXX (The number assigned by the OpenConf System).
[33]. Wibowo, H., Suripin, Kodoatie, R., and Isdiyana., 2015.” Comparingthe Calculation Method of the
Manning Roughness Coefficientin Open Channels”.Proceedings of International Conference,The 1st
Young Scientist International Conference of Water Resources Development and Environmental
Protection, Malang, Indonesia.
[34]. Yang, S.Q., and Tan, S.K., 2008. “ Flow Resistance over Mobile Bed in an Open-Channel Flow
“,Journal of Hydraulic Engineering, ASCE, Vol. 134,No. 7,pp. 937-947.
[35]. Yanti., N., 2011.” Application of Neural Network Methodwith Back Propagation Structure for Drug
Stock Prediction At Pharmacies (Case Study: Pharmacies ABC), National Seminaron Information
Technology Applications (SNATI 2011) ISSN: 1907-5022, Yogyakarta.
[36]. Yen, B. C., 2002. “ Open Channel Flow Resistance.” Journal of Hydraulic Engineering” , Vol 128(1).
Pp. 20-39.
[37]. Yen, B. C.,1992. „„Dimensionally Homogeneous Manning‟s Formula.‟‟ J.Hydraul. Eng.,118-(9), 1326
–1332; Closure: (1993),.119(12), 1443–1445.
a.

Application Methods artificial neural network(Ann) Back propagation structure for Predicting the value Of bed channel roughnes scoefficient

More Related Content

What's hot (19)

Viewers also liked (8)

Similar to Application Methods artificial neural network(Ann) Back propagation structure for Predicting the value Of bed channel roughnes scoefficient (20)

More from irjes (20)

Recently uploaded (20)

Application Methods artificial neural network(Ann) Back propagation structure for Predicting the value Of bed channel roughnes scoefficient