Document Type : Research Article

Authors

Lorestan University

Abstract

Introduction: Forecasting and modeling of river flow is an essential step towards planning, designing and utilizing water resources management system which is subject to issues such as droughts and destructive floods in river basins. The river flow deficit and excess could result in financial and human losses. Such predictions of river flow not only provide the necessary warning signals about the flood risk, but also help to adjust the water outflows during low level of water flows which help to the water resource management. Due to the importance of river flows and its fluctuations in short and long term on different aspects of human lives, understanding its behavior and performance is crucial (necessary). Thus, with discovering its dynamic behavior, it is possible to predict its future performance. The aim of this study is to explore and simulation of Kashkan River’s performance using the statistical intelligent methods to provide models with lower uncertainty in order to improve the planes based on Kashkan’s River flows.
Materials and Methods: For this study, the series of daily discharge data from Poldokhtar- Kashkan station (located in the coastal river) over 1370-1393 were used as the primary input. Methods used in this study were based on memory uses the Hurst exponent of long memory time series. Runoff is the dynamics of the series. The current state of these series is dependent on its historical states. The delay time (lag time) of 1, 3, 5, 7, 10 and 15 days before the runoff were calculated. The amount of runoff was seen as a function of the time series. Considering the above-mentioned six time series as input signals, time series modeling using statistical methods K- nearest neighbor (K-NN), and artificial neural network, combined wavelet - K-NN and combining the wavelet nervous.
Results and Discussion: Kashkan’s Memory river flow system, using the Hurst exponent within 10 days and mid-4200 based on the amount of 0.6 was obtained (Figure 2). This amount indicates a non-linearly behavior and a dynamic learning system. In addition, it shows the presence of long memory in the river flow time series. Then, by allocating 80% of the data for training and the remaining 20 percent for testing the model and adopting ranges from 1 to 10 nearest neighbor and a range of 1,000 to 50,000 particles (for data on education) Model K-NN were prepared. Using the criteria to assess the efficiency and accuracy of a model in each performance of the mentioned domains, the best model with the 6 neighbors structure and 15,000, was obtained. In this model stimulated the runoff with the correlation of 0.90 and a 4.6 error was obtaied. On the other hand, artificial neural network architecture to simulate runoff with 6 input neurons in a hidden layer neurons and considering 3 to 20 and an output neurons leading to the 6-8-1 structure as the best model was fitted. This model has a correlation of 0.89 and the forecast error of 5.8 in the process of runoff simulation. Then using wavelet function, mortality, time-series signal runoff into 4 levels, including 8 under high frequency and low frequency signal was decomposed where high-frequency signals and low-frequency signal of 4 level were considered as the original signal for the input surface runoff. In this regard, the hybrid model K-NN-WT with runoff time series prediction error of 2.7 percent and the hybrid model ANN-WT with the correlation of 0.99 the estimation error of 1.2 were simulated.
Conclusion: Running 4 Artificial Neural Network (ANN), K-nearest neighbor (K-NN) and combining the wavelet analysis of the two models (ANN-WT and K-NN-WT) to predict the time series of runoff river showed that due to the existence of multiple time frequencies in the time series of the river signals, its decomposition it using wavelet analysis results in extraction of hidden information that are not available through the original signal. This information is the daily, monthly, quarterly and annual fluctuations. The hybrid models performance indicated higher accuracy and improved outcomes relative to individual models. In fact, the analysis of the original runoff signal by wavelet analysis in the process of simulation results in an appropriate weighting given to long-term and short term dynamic of runoff which led to significant lower error in modeling

Keywords

1- Abdollahi Asadabadi S., Dinpashoh Y., and Mirabbasi R. 2014. Forecasting of mean daily runoff discharge of behesht-abad River using wavelet analysis. Journal of Water and Soil, 28(3):534-545. (in Persian with English abstract)
2- Akhtar M.K., Corzo G.A., Van Andel S.J., and Jonoski A. 2009. River flow forecasting with artificial neural networks using satellite observed precipitation pre-processed with flow length and travel time information: case study of the Ganges river basin. Hydrol. Earth Syst. Sci, 13:1607–1618.
3- Anis Hosseini M., and Zaker Mashgh M. 2013. Analysis and forecasting of river flow kashkan using chaos theory. journal of hydrolic, 8(3):45-61. (in Persian with English abstract)
4- Azmi M., and Araghinejad. 2012. Development of K-Nearest Neighbour regression method in forecasting river stream flow. J. of Water and Wastewater, 2:108-119. (in Persian with English abstract)
5- Cannas B., Fanni A., See L., and Sias G. 2006. Data preprocessing for river flow forecasting using neural networks: Wavelet transforms and data partitioning. Phys. Chem. Earth, 31(18): 1164-1171.
6- Daubechies I. 1992. Ten lectures on wavelets. Society for Industrial Mathematics.
7- Haghizadeh A., Mohammadlo M., and Nouri F. 2015. Modeling rainfall – runoff process using artificial neural network and Neuro-Fuzzy Computing and multiple regression (case study: watershed of Korramabad). journal of Eco hydrology, 2:233-243. (in Persian with English abstract)
8- Hassanzadeh Y., Lotfollahi M.A., Shahverdi S., Farzin S., and Farzin N. 2013. De-noising and prediction of time series based on the wavelet algorithm and chaos theory (Case Study: SPI drought monitoring index of tabriz city), Iran-Water Resources Research, 8(3):1-13. (in Persian with English abstract)
9- Hurst H.E .1951. Long-term storage capacity of reservoirs (with discussion). Transactions of the American Society of Civil Engineers, 116: 770–808.
10- Karamuz M., and Araghinejad Sh. 2014. Advanced hydrology. AmirKabir University. Iran
11- Kia, M. 2010. Neural networks in matlab. Qian academic publishing.
12- Kisi O. 2007. Streamflow forecasting using different artificial neural network algorithms. ASCE Journal of Hydrologic Engineering, 12(5):532-539.
13- Kisi O. 2005. Daily river flow forecasting using artificial neural networks and auto-regressive models. Turkish J. Eng. Env. Sci, 29:9-20.
14- Lall U., and Sharma A. 1996. A nearest neighbor bootstrap for resampling hydrologic time series. Water Resources Research, 32(3):679-694.
15- Lee S., Ryu J.H., Lee M.J., and Won J.S. 2006. The Application of artificial neural networks to landslide susceptibility mapping at Janghung, Korea. Mathematical Geology, 38(2):199-220.
16- Mark H.B., Martin T.H., and Haward B.D. 2016. Neural network tolboxTM getting started guide. The MathWorks, Inc.
17- Menhaj M. 2002. Neural networks and artificial intelligent basic. First edition AmirKabiruniversity. Press, 350p.
18- Montaseri M., and Zamanzad Ghavidel S. 2014. River Flow Forecasting by Using Soft computing Journal of Water and Soil, 28 (2):394-405. (in Persian with English abstract)
19- Nayak P.C., Sudheer K.P., Rangan D.M., and Ramasastri K.S. 2004. A neuro-fuzzy computing technique for modeling hydrological time series. Journal of Hydrology, 291(1):52-66.
20- Pramanik N., and Panda R.K. 2009. Application of neural network and adaptive neuro-fuzzy inference systems for river flow prediction. Hydrological sciences journal, 54(2): 247-260.
21- Sanikhani H., Dinpashoh Y., and Ghorbani M.A. 2014. Baranduz-chay river flow modeling using the K-nearest neighbor and intelligent methods. water and soil science, 25(1):219-233.
22- Shafaei M., Fakheifard A., Darbandi S., and Ghorbani M.A. 2013. predicrion daily flow of vanyar station using ANN and wavelet hybrid procedure. Irrigation & Water Engineering, 14:144-128. (in Persian with English abstract)
23- Shataee Sh., Kalbi S., Fallah A., and Pelz D. 2012. Forest attributes imputation using machine-learning methods and ASTER data: comparison of k-NN, SVR and random forest regression algorithms. International Journal of Remote Sensing, 33(19):6254-6280. (in Persian with English abstract)
24- Veiga V.B., Hassan Q.K., and He J. 2015. Development of Flow Forecasting Models in the Bow River at Calgary, Alberta, Canada. journal Water, 7:99-115.
25- Wang W., and Ding J. 2003. Wavelet network model and its application to the prediction of hydrology. Nature and Science, 1(1):67-71
26- Wilson D. R., and Martinez T. R. 2000. Reduction techniques for exemplar-based learning algorithms. Machine Learning, 38(3): 257-286.
27- Wu C.L., and Chau K.W. 2010. Data-driven models for monthly streamflow time series prediction. Engineering Applications of Artificial Intelligence 23:1350-1367.
28- Yates D., Gangopadhyay S., Rajagopalan B., and Strzepek K. 2003. A technique for generating regional climate scenarios using a nearest-neighbor algorithm. Water Resoures Research, 39 (7): 1114- 1121.
29- Young C.C., Liu W.C., and Chung C.E. 2015. Genetic algorithm and fuzzy neural networks combined with the hydrological modeling system for forecasting watershed runoff discharge. The Natural Computing Applications, 1-13.
CAPTCHA Image