Inflow modelling

8/11/2019 Inflow modelling

1/37

1

Inflow modeling of Okhla barrage using

ANN and Fuzzy- based models

Final Presentation29thNovember,2011

Vipin Sharma (08131)Dhruv Chaudhary (08109)

Amit Sharma (08140)

Amit Sehara (08144)

Abhitosh Yadav (08142)

Department of Civil Engineering

National Institute of Technology,

Hamirpur (H.P.)

SupervisorDr. V.S.Dogra


2/37

2

Outline of Presentation

Introduction

Objectives of the study

Reservoir under study

Methodology

Model Development

Results

Comparisons

Conclusions

References


3/37


4/37

Statistical Models

4

Data driven models.

Result in better and comparatively easier simulation for non- linear

relationships.

In recent decades, many researchers have shown a great deal of interest in

statistical / mathematical modeling of various hydrological processes.

Statistical models have performed better then the traditional methods in

modeling various hydrological processes like sediment-discharge

relationships. (Sarkar,2010)


5/37

Objectives of Study

To understand the method of modeling a physical process using the

technique of fuzzy-based model and Artificial Neural Network.

Application of these techniques for inflow modeling of Okhla barrage

using previous yearsdata.

Comparison of efficiency and accuracy of ANN model and fuzzy-

based model.

5


6/37

Okhla Barrage

Constructed on Yamuna river

Situated at the border of U.P. and Delhi.

Constructed in 1986.

Constructed and maintained by Department of Irrigation, U.P.

6


7/37


8/37

8

Upstream daily inflow data of years 2000 2010 was received from

Department of Irrigation U.P.

The units of data is cubic feet per second.

This data was then processed using Microsoft Excel .

70% of this data was used for the development of models and rest 30% for

validation and testing.

DATA


9/37


10/37

Model development

Two types of models have been developed for prediction of inflow.

1. Short term prediction

2. Long term Prediction

Short term prediction

These models have been developed for short term prediction of inflow.The models have been developed to predict the inflow:

One day in advance: (e.g. inflow on May 15th, 2011 can be predicted on May

14th, 2011);

Three days in advance: (e.g. inflow on May 15th, 2011 can be predicted on

May 12th, 2011);

Seven days in advance(e.g. inflow on May 15th, 2011 can be predicted on

May 8th, 2011);

10


11/37

Model development

Long term Models

These models have been developed to predict the inflow one year in advance.

(e.g. inflow on May 15th, 2011 can be predicted on May 15th, 2010).

A total number of seven models have been developed using ANN for this

purpose. Similarly, 3 models have been developed using Subtractive

clustering cum fuzzy if-then rule base.

11


12/37

Model Names

Long term Models: Named on the basis of number of inputs and the method

of Development.

e.g. ANN3 is the model made by ANN using 3 inputs.

FSC-3 is the model made by fuzzy and subtractive clustering using 3 inputs.

Short Term Models: Named on the basis of prediction lag and no. of inputs.

Eg. TS12 is short term prediction model made by time series tool

which can predict inflow 1 day in advance and uses 2 inputs.

12


13/37

Inputs

Past inflow values has been used as input(independent variable) for

prediction of inflow at given time(dependent variable).

These input parameters were selected by studying visible patterns in

the data or on hit and trail basis.

Different models were developed using variable no. of input

parameters and variable no. of hidden layers(ANN).

Campolo et al. (1999) suggested that the capacity of a model to

generalize inflow is more accurate when discharge values are used as

input to the model. 13


14/37

Inputs

ANN1: The input which is used for this model is the 1 year lagged inflow

value. E.g. for prediction of inflow on 15thMay, 2010, input- 15thMay, 2009.

ANN 2: In ANN 2, the inputs used are 1-year and 2- year lagged inflow

values. E.g. to predict the inflow on 15th

May 2010, the inputs -15th

May,

2009 and 15thMay, 2008.

ANN 3: In ANN 3 model, the inputs are 1 year, 2 year and 3 year lagged

inflow values. E.g. To predict the inflow of 15thMay 2010,the inputs 15th

May 2009, 15thMay 2008 and 15thMay 2007.

ANN 4: In ANN4, the inputs are 1-year, 2-year, 3-year and 4-year lagged

inflow values. E.g. To predict the inflow of 15thMay 2010, the inputs - 15th

May 2009, 15thMay 2008 , 15thMay 2007 and 15thMay 2006.

14


15/37

Inputs

5. ANN 5:In this model, the inputs are 1year, 2year, 3year, 4year lagged inflow

values and monthly average inflow value of corresponding month lagged by 1

year. E.g. the prediction of inflow on 15thMay, 2010 ,inputs-15thMay, 2009,

15thMay, 2008, 15thMay 2007, 15thMay 2006 and average inflow value of

May, 2009.

6. ANN 8:In this model, the inputs are 1year, 2year, 3year, 4year lagged inflow

values and monthly average inflow value of corresponding month lagged by 1

year, 2year, 3year and 4year.e.g. The prediction of inflow on 15 thMay 2010

depends on the inflow values of 15thMay 2009, 15thMay 2008, 15thMay 2007,

15thMay 2006 and average inflow value of May 2009, May 2008, May 2007

and May 2006.15


16/37

Inputs

7. ANN 12:In this model, the inputs are 1-year, 2-year, 3-year and 4-year

lagged inflow values of same day, monthly average inflow value of

corresponding month lagged by 1-year, 2-year, 3-year and 4-year and annual

average inflow value lagged by 1-year, 2-year, 3-year and 4-year.

E.g. the prediction of inflow on 15thMay 2010 depends on the inflow values of

15thMay 2009, 15thMay 2008, 15thMay 2007, 15thMay 2006 plus average

inflow value of May 2009, May 2008, May 2007 and May 2006 and average

inflow value of 2009, 2008, 2007 and 2006.

16


17/37

Inputs

Fuzzy based Models

The inputs used for the fuzzy models were same as that of corresponding

model of ANN. i.e. the inputs for FSC-3 were same as that of ANN3 and so

on.

17


18/37

EVALUATION CRITERIA

18


19/37

EVALUATION CRITERIA

4. Regression coefficient: Directly obtained using Neural- fitting tool box.

5. Pearson Correlation coefficient (R): It gives a measure of correlation

between the Actual and Predicted inflow dataset. Its value lies between -1

to +1. Values near to zero depicts poor model Performance. Its value is

calculated as follows.

Where Xiis the observed inflow, Yiis the predicted in flow.

19


20/37

ANN12 : Regression Values

20


21/37

RESULTS- ANN Long Term Models

Model NRMSE % RMSE R CE Rc

ANN1 .204 33.22 0.14 0.05 0.288

ANN2 0.193 31.5 0.30 0.05 0.39

ANN3 0.17 27.8 0.55 0.26 0.56

ANN4 0.17 27.0 0.56 0.22 0.61

ANN5 0.15 24.8 0.66 0.41 0.71

ANN8 0.11 18 .4 0.83 0.68 0.88

ANN12 0.10 19.1 0.84 0.70 0.89

21

Table 5.1 Performance criteria for long term prediction ANN models


22/37

ANN12- Predicted Inflow

22


23/37

RESULTS-ANN short term model

Model NRMSE % RMSE R CE

TS12 0.12 23.5 0.92 0.86

TS13 0.07 13.4 0.93 0.86

TS15 0.07 13.9 0.92 0.85

TS17 0.07 13.2 0.93 0.86

TS32 0.11 22.7 0.80 0.61

TS33 0.11 20.1 0.83 0.69

TS34 0.11 20.7 0.82 0.67

TS37 0.11 21.1 0.83 0.66

TS72 0.20 39.7 0.39 0.21

TS77 0.14 27.2 0.68 0.43

23

Table 5.2 Performance criteria for short-term prediction ANN models


24/37


25/37

TS17- Time Series Prediction

25


26/37

Results- Fuzzy models


FSC-3 0.24 25.3 0.25 0.56

FSC-5 0.17 19.3 0.64 0.40

FSC-8 0.15 15.7 0.84 0.59

26

Table 5.3 Performance criteria for fuzzy based models


27/37

FSC-8 FIS- Editor

27


28/37

FSC-8 Final Model

28


29/37

Comparisons

1. Hidden Layers: this comparison was done on model ANN4

29


30/37

Comparisons

30

Training Algorithm: Comparison done on Model ANN3


31/37

Comparison: ANN long term prediction model

31


32/37

ANN vs. Fuzzy

32


FSC-3 0.24 25.3 0.25 -0.56

ANN3 0.17 27.8 0.55 0.26


33/37

ANN vs. Fuzzy

33

Performance of ANN 3 and FSC-3 in modelling inflow into barrage

for december,2010


34/37

Conclusion

Long term Prediction model

Model ANN12 was found in terms of performance with regression coefficient

value as high as 0.93 during training.

The Performance of model also increases with increase in number of neurons in

hidden layer, but it tends to increase the processing time for training.

The optimum number of neurons was found to be in range of 100 during this

study.

The fuzzy and ANN models do not show much of difference in the performancefor same set of inputs

once the number of inputs was increased beyond eight, the fuzzy based models

started giving incoherent results.34


35/37

Conclusion

Short term Prediction Models

The accuracy of these models was found out to be relatively higher than

that of long term prediction models

The accuracy of one day lagged models was found to be maximum with

regression value greater than 0.90.

Accuracy of prediction models decreased as the Lag was increased.

35


36/37

36

References

Subribabu ,C.R., Ramayya,V.V. and Kumar,R.P. (2005). Application of

Artificial Neural Network Model for Inflow Prediction,Journal of Indian Water

Resources Society,Vol.35, No.4, PP. 9-24.

Sarkar, A., Raju ,M.M. and Kumar,A. (2010). Sediment Runoff Modelling

Using Artificial Neural Networks, Journal of Indian Water Resources SocietyVol.30, No.1, PP. 39-45.

Sarkar, A., Agarwal, A. and Singh, R.D. (2006). Artificial Neural Networks

Models for Rainfall Runoff Forecasting In a Hilly Catchment, Journal of

Indian Water Resources Society , Vol.26, No.2, PP. 5-12.

Adamowski, J. and Karapatki, C. (2010). Comparison of Multivariate

Regression and Artificial Neural Networks for Peak Urban Water Demand

Forecasting: Evaluation of Different ANN Learning Algorithm,ASCE Journal of

Hydrological Engineering , Vol.15, No.10, PP. 729-743.


37/37

37

EI-Shafie, A., Taha, M.R. and Noureldin ,A. (2007). A neuro-fuzzy model for

inflow forecasting of the Nile river at Aswan high dam,Water Resource Manage,

Vol.21, PP. 533-556.

Deka, P.C. and Chandramouli,V. (2009). Fuzzy Neural Network Modeling of

Reservoir Operation, ASCE Journal of Water Resources Planning and

Management, Vol.135, No.1,PP. 5-12.

Googhari, S.K., Feng, H.Y., Ghazali ,A.H. and Shui, L.T. (2010). Neural

Network for Forecasting Daily Reservoir Inflows, Pertanik J.Science &

Technology, Vol.18,No.1, PP. 33-41.

Othman, F. and Naseri, M. (2011). Reservoir inflow forecasting using artificial

neural network, International Journal of the Physical Science,

Vol.6,No.3,PP.434-440.

References

Inflow modelling

Documents

Transcript of Inflow modelling