مدل سازی تخمین میزان رسوب رودخانه به کمک روش شبکه عصبی مصنوعی (نمونه موردی: رودخانه گلرود)

نوع مقاله: مقاله پژوهشی

نویسندگان

1 دانشگاه محقق اردبیلی

2 دانشگاه محقق ردبیلی

چکیده

رسوبات رودخانه ای به دو صورت منتقل میشوند: یا این مواد درون جریان آب غوطه ور هستند و همراه با آب در حرکت می باشند که به آنها مواد رسوبی معلق گفته میشود و میزان مواد رسوبی معلق را که در واحد زمان از یک مقطع رودخانه عبور کند، بار معلق مینامند، یا اینکه به یکی از صور لغزش، غلتیدن، پرش حرکت مینمایند که به آنها بار بستر می گویند. شبکه عصبی مصنوعی روشی است که بر پایه شبیه سازی عملکرد مغز انسان بـرای حـل مـسایل متنوع ارایه و از لایه های نرون ورودی، خروجی و میانی و وزنهای مربوط به مقادیر ورودی و بایاس و تابع تحریک تشکیل شده است. منطقه مورد مطالعه در این پژوهش حوضه آبریز رودخانه گِلِرود است. این منطقه در شهرستان بروجرد، در استان لرستان در غرب ایران واقع شده است، پژوهش حاضرازنوع کاربردی ست. بدین صورت که، ابتدا مشخصات زیرحوضه های این رودخانه استخراج شده است این مشخصات شامل مشخصات فیزیکی زیرحوضه ها از جمله مساحت، محیط و طول آبراهه ها و مشخصات مربوط به دبی رودخانه و میزان رسوب آن است. در ادامه با روش های رگرسیون خطی چند متغیره، شبکه عصبی پیش خور چندلایه (MLP) و شبکه عصبی برپایه تابع شعاعی (RBF) به مدل سازی تخمین رسوب پرداخته شده است.پس از محاسبه شاخص های RMSE و MAE با توجه به این امر که هرچقدر میزان این شاخص ها کمتر باشد مقدار پیش بینی شده به مقادیر واقعی نزدیکتر است بنابراین باتوجه به شواهد حاصله مدل شبکه عصبی مصنوعی MLP دقت بهتری را نسبت به دو مدل دیگر در تخمین میزان رسوب منطقه نشان میدهد. از سوی دیگر با توجه به مقدار شاخص R2 که برای سه مدل محاسبه شده است دقت تخمین مدل به مقدار 0.409 برای مدل MLP محاسبه شده است، مقدار R2 برای این مدل برابر 0.88 است. پس از مدل شبکه عصبی مصنوعی MLP، مدل شبکه مصنوعی RBF نتایج بهتری ارائه می دهد. در این مدل مقدار R2 برابر است با 0.4 که نشان دهنده دقت تخمین حدود نصف مدل MLP است. و در رتبه سوم نیز مدل رگرسیون خطی چند متغیره با مقدار R2 برابر با 0.3 قرار دارد.مدل رگرسیون خطی نیز به علت این امر که تنها روابط خطی بین متغیر ها را در نظر می­گیرد دارد بیشترین میزان خطا است.

کلیدواژه‌ها


عنوان مقاله [English]

Modeling the estimation of river sediment with the help ofartificial neural network method (Case study: Geleroodriver)

نویسندگان [English]

  • Dariush Abolfathi 1
  • Aghil Madadi 2
  • Sayyad Asghari 1
چکیده [English]

    Introduction
River sediments are transmitted in two ways: either these substances are immersed in the flow of water, and they move with water, which is called suspended sediment, and the amount of suspended sediment that passes through a section of the river at a time They call a suspended load, or they move a slip, slide, jump, to which they say the bed load. Artificial neural network is a method that is based on the simulation of human brain function for solving various problems and the input, output, and median of the neuron layers and the weights associated with the input values ​​and the bias and the stimulation function. The study area in this study is the catchment area of ​​Golrood River.
Artificial neural network is a method whichwasprovided based on the simulation of human brain function for solving various problems and formed from the input, output, and median neuron layers and the weights associated with the input values and the bias and the stimulation function. One of the features of the artificial neural network can be referred to as the calculation of a definite function, the approximation of an unknown mapping, pattern recognition, signal processing, and learning (American Society of Civil Engineers, 2000). The disadvantages of neural network methods are that it does not provide a function which can be used explicitly. Many studies have not been conducted on sedimentation using a neural network (Govindaraju&Ramachandra, 2000; Sarangi & Bhattacharya, 2005).Feedforward error back propagation neural networks with nonlinear functions (sigmoid) has high flexibility and can be very effective in approximating a function, finding the relation between input and output, and so on. In hydrology, the use of these networks is highly recommended considering the turbulence dominating runoff-sediment data, (Flood &Kartam, 1994).
Javadi and others (2015) in an article compared river sediment estimation method using two methods of artificial neural network and SVM in Iran. Then, the output of these models was compared with the experimental models and eventually the RMSE and R indices to compare these models were used. The results indicated that SVM model has better estimation than artificial neural network model. The RMSE was 75 for this model.
Semkol et al. (2016) estimated the amount of sediment in the Shiwan River in Taiwan. In this study, artificial neural network model and sediment rating curve method were used. The results showed that the MLP neural network model was able to provide an appropriate estimation of the amount of sediment with R value of 0.97 (Tfwala& Wang, 2016).Afan et al. (2016) also estimated the amount of river sediment in the Johouw River. In this study, two models of neural network RBF and FFNN were used. Finally, it was found that the FFNN model showed a much better performance than the RBF model. The R index of this study for the FFNN model was 0.92 and its RMSE was 26, while the RBF model had R value of 0.86 and a RMSE of 32 (Afan et al., 2015).
Summarizing the research history showed that the static regression methods did not have high accuracy in estimating the suspended sediment load discharge.In recent years, the focus of predictive models has also changed from linear regression to neural network models. Most researchers have been providing comparisons between different models of the neural network during these years and also, in their final modeling, tried to use the domain morphology factors in the final model to improve the accuracy of the final model. Therefore, the use of artificial neural network method and considering the dynamic behavior of sediment suspension load and considering the flow of previous days as an effective variable has been evaluated in this research.
 
 Materials & methods
 The study area
The study area in this research is the Gelerood river basin. This area is located in Borujerd, Lorestan province, west of Iran. The basin is between longitudes 48.30 to 48.55 degrees and latitudes 33.45 to 34.00 degrees. GeleroodRiver drains waters of an area of 70 square kilometers. The average height of this basin is 2350 meters. The river originates from a number of headwaters in the village of Vanai in the west of this city and receives other branch in the western part of the Boroujerd city in the vicinity of the Chogha hill from the north.
There are 8 stations named such as Doroud-Tireh, Doroud-Marbareh, DarehTakht-Marbareh, Vanai-Gelerood, Biatoun, Rahim-Abad, Water Organization and Chogha hill in the area of Silakhor plain in Dorood-Boroujerd area. In Gelerood river basin, two stations of Vanai and Water Organization have been used to estimate the amount of river sediment. The position of these two stations in relation to the Gelerood River and its sub-basins is shown in Fig 1.
 Data used
In this study, the instantaneous flow rate- instantaneous sediment statistics recorded deposition related to the period 1971 to 2002 were used. These figures include the instantaneous daily flow rate per cubic meter per second and the instantaneous daily sediment per day that were measured simultaneously. Morphological characteristics of the basin including the area, length of the river and its environment using ArcGIS software and geomorphologic parameters of the basin using natural features of the basin have been calculated based on the guidelines of Singh et al. (2009) using the ArcHydro plugin installed on the above software.
 
 Results
So far, different prediction models have been used to estimate the sediment volume of rivers. Some of these models estimated the amount of sediment by combining various physical parameters of the domain, climate, and even satellite image outputs. Artificial neural network models are widely used today to predict geographic models. In this study, three models of artificial neural network RBF, artificial neural network MLP and multivariate linear regression model have been used to estimate river sediment.
After calculating the RMSE and MAE indices, given the lower the rate of these indicators, the predicted value is closer to the actual values, so MLP artificial neural network models have a better accuracy than the other two other models in estimating the region's sediment. On the other hand, according to the value of the  index calculated for the three models, the accuracy of the model estimation is calculated 90.44 for the MLP model, the  value for this model is 0.88. After the MLP artificial neural network model, the RBF artificial network model provides better results. In this model, the value of  is 0.4, which indicates the estimate accuracy of the half of the MLP model. In the third place, the multivariate linear regression model with  value is 0.3.
Two neural network models of MLP and RBF were also studied in this research. The MLP model was able to estimate sediment data with a better accuracy than other models. Thus, the feasibility of using feedforward neural network models in the estimation of sediment load can be confirmed. Based on the available time series, more accurate estimates require long periods of time, as well as considering climate changes in this research can help improve the results and accurately predict the amount of sediment. On the other hand, taking into account the soil type-specific parameters of the area and the potential for water penetration in the soil for each sub-basin can be effective in improving the results. The results of this study indicated that there is a significant relationship between the amount of suspended sediment production with the number and severity of runoff events. Among the physical characteristics, the area of the basin and the length of the main river are other factors that affect the estimation of the river downstream sedimentation rate.
As well as, recurrent neural network models can be used in the following studies, given that the stations are located along the other stations. Moreover, the combination of satellite imagery data can lead to more accurate models, given the fact that this data is also available to users from past periods.

کلیدواژه‌ها [English]

  • Gelerood
  • Artificial Neural Network
  • estimation of sediment
  • liner regration
  • MLP
  • RBF
  • اکبرپور, ا., & حامدافتخار, خ. (1385). مقایسه مدلهای شبکه عصبی مصنوعی و رگرسیون در پیش بینی آورد رسوب درحوضه اهرچای آذربایجان شرقی. اولین همایش ملی مدیریت شبکه های آبیاری و زهکشی، دانشگاه شهید چمران اهواز، دانشکده مهندسی علوم آب.
  • ولی, ع., معیری, م., رامشت, م., & موحدی نیا, ن. (1389). حلیل مقایسه عملکرد شبکه های عصبی مصنوعی و مدل های رگرسیونی پیش بینی رسوب معلق مطالعه موردی: حوضه آبخیز اسکندری واقع در حوضه آبریز زاینده رود. پژوهشهای جغرافیای طبیعی (پژوهش های جغرافیایی), 71(42), 21–30. Retrieved from http://fa.journals.sid.ir/ViewPaper.aspx?ID=105781
  • شفاعی بجستان م، 1373. هیدرولیک رسوب. انتشارات دانشگاه شهید چمران اهواز.
  • همایون فقیه و همکاران 1394 بررسی کارایی شبکه عصبی مصنوعی در برآورد بار معلق رودخانه با استفاده از داده های دسته بندی شده
  • علیرضا مردوخ پور1396 ارزیابی میزان برآورد رسوب با بهره گیری از روش منحنی سنجه و مقایسه نتایج با روش های رگرسیونی و شبکه عصبی
  • محمد طهمورث 1387 مقایسه دقت مدل های شبکه عصبی مصنوعی ژیومرفولوژی(GNNS) و رگرسیونی(RM) در برآورد رسوب طالقان رود. پژوهش های آبخیزداری
  • Abolvaset, N., and Shahradfar, S. 2006. Investigation the effect of river water level oscillation on suspended sediment using Artificial Neural Networks (Application in the Ahar River Watershed in Satarkhan Dam). 7th International River Engineering conference. Shahid Chamran University, Ahwaz, Iran, Pp: 235-243.
  • Avarideh, F., Banihabib, M., and Tahershamsi, A. 2001. Application of ANN for Estimation of Sediment Load in Rivers. 3rd Iranian Hydraulic Conference,Tehran University, Iran, Pp: 178-186. (In Persian)
  • Afan, H. A., El-Shafie, A., Yaseen, Z. M., Hameed, M. M., Wan Mohtar, W. H. M., & Hussain, A. (2015). ANN Based Sediment Prediction Model Utilizing Different Input Scenarios. Water Resources Management, 29(4), 1231–1245. https://doi.org/10.1007/s11269-014-0870-1
  • Demir, G., Ozdemir, H., Ozcan, H. K., Ucan, O. N., & Bayat, C. (2010). an artificial neural network-based model for short-term predictions of daily mean PM10 concentrations. Journal of Environmental Protection and Ecology, 11(3), 1163–1171.
  • Doxaran, D., Froidefond, J. M., Castaing, P., & Babin, M. (2009). Dynamics of the turbidity maximum zone in a macrotidal estuary (the Gironde, France): Observations from field and MODIS satellite data. Estuarine, Coastal and Shelf Science, 81(3), 321–332. https://doi.org/10.1016/j.ecss.2008.11.013
  • Flood, I., & Kartam, N. (1994). Neural Networks in Civil Engineering. I: Principles and Understanding. Journal of Computing in Civil Engineering, 8(2), 131–148. https://doi.org/10.1061/(ASCE)0887-3801(1994)8:2(131)
  • Gong, B., Im, J., & Mountrakis, G. (2011). An artificial immune network approach to multi-sensor land use/land cover classification. Remote Sensing of Environment, 115(2), 600–614. https://doi.org/10.1016/j.rse.2010.10.005
  • Govindaraju, R. S., & Ramachandra, A. (2000). Artificial Neural Networks in Hydrology. Artificial Neural Networks in Hydrology, 36(April), 337. https://doi.org/10.1007/978-94-015-9341-0
  • Güneralp, İ., Filippi, A. M., Hales, B. U., I, G., M, F. A., & U, H. B. (2013). River-flow boundary delineation from digital aerial photography and ancillary images using Support Vector Machines. GIScience & Remote Sensing, 50(1), 1.
  • https://doi.org/10.1080/15481603.2013.778560
  • Lopes, V. L., Ffolliott, P. F., & Baker, M. B. (2001). Impacts of Vegetative Practices on Suspended Sediment from Watersheds of Arizona. Journal of Water Resources Planning and Management, 127(1), 41–47. https://doi.org/10.1061/(ASCE)0733-9496(2001)127:1(41)
  • M, P. T., & C, S. L. (2008). Rivwidth: A Software Tool for the Calculation of River Widths from Remotely Sensed Imagery. IEEE Geoscience and Remote Sensing Letters, 5, 70.
  • Mirbagheri, S., and Rajaei, T. 2006. Improvement of suspended load prediction by artificial neural networks. 7th International Civil Engineering Conference.Tarbiat Modares Univ., Tehran, Iran, 2006, Pp: 435-443. (In Persian)
  • Mirbagheri, S.A., and Rajaei, T. 2004. Use of ANN in Estimation of Suspended Sediment Load of Zohre River. 1st International Conference of Civil Engineering, Tehran University, Iran, 2004, Pp: 45-53. (In Persian)
  • M, M. V. (2007). An Automated GIS Procedure for Delineating River and Lake Boundaries. Transactions in GIS, 11, 213.
  • Nakato, T. (1990). Tests of Selected SedimentTransport Formulas. Journal of Hydraulic Engineering, 116(3), 362–379. https://doi.org/10.1061/(ASCE)0733-9429(1990)116:3(362)
  • Rajaei, T., Mirbagheri, S.A., Bodaghpour, S., and Zoneamat Kermani, M. 2007.Use of ANN in order of Modeling of Nonlinear Time series of Suspended Sediment Load in Rivers. 6th Iranian Hydraulic Conference. Shahrekord University, Iran. (In Persian)
  • Sarangi, A., & Bhattacharya, A. K. (2005). Comparison of Artificial Neural Network and regression models for sediment loss prediction from Banha watershed in India. Agricultural Water Management, 78(3), 195–208. https://doi.org/10.1016/j.agwat.2005.02.001
  • SPSS Inc. (2007). SPSS Advanced Statistics 17.0. Statistics, 189.
  • Tfwala, S. S., & Wang, Y. M. (2016). Estimating sediment discharge using sediment rating curves and artificial neural networks in the Shiwen River, Taiwan. Water (Switzerland), 8(2). https://doi.org/10.3390/w8020053
  • Walling, D. E., & Webb, B. W. (1988). The reliability of rating curve estimates of suspended sediment yield: some further comments. Sediment Budgets (Proceedings of the Porto Alegre Symposium), (174), 337–350.
  • Zhu, Y.M., Lu, X.X., and Zhou, Y. 2007. Suspended sediment flux modeling with artificial neural network: An example of the Longchuanjiang River in the Upper Yangtze Catchment, China. Geomorphology, 84: 111-125.