تخمين جريان روزانه رودخانه با استفاده از مدل‌هاي هوشمند، مطالعه موردي: رودخانه مهاباد

Daily river flow estimation based on intelligent models, case study: Mahabad River

ﺗﺨﻤﯿﻦ ﺻﺤﯿﺢ و دﻗﯿﻖ ﺟﺮﯾﺎن رودﺧﺎﻧﻪ ﻣﯽﺗﻮاﻧﺪ ﻧﻘﺶ ﻣﻬﻤﯽ در ﮐﺎﻫﺶ اﺛﺮات ﻧﺎﺷﯽ از ﺧﺴﺎرات ﺳﯿﻼب اﯾﻔﺎ ﮐﻨﺪ. در اﯾﻦ ﺗﺤﻘﯿﻖ، از ﻣﺪل ﺑﺮﻧﺎﻣﻪرﯾﺰي ﺑﯿﺎنژن )GEP( و ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ )BN( ﺑﺮاي ﭘﯿﺶﺑﯿﻨﯽ ﺟﺮﯾﺎن روزاﻧﻪ رودﺧﺎﻧﻪ ﻣﻬﺎﺑﺎد واﻗﻊ در ﺣﻮزه آﺑﺨﯿﺰ درﯾﺎﭼﻪ اروﻣﯿﻪ اﺳﺘﻔﺎده ﺷﺪ. ﺑﺮ اﯾﻦ اﺳﺎس، از ﭼﻬﺎر اﻟﮕﻮي ورودي ﺑﺎ ﺗﺄﺧﯿﺮﻫﺎي ﯾﮏ ﺗﺎ ﭼﻬﺎر روزه ﺑﺮاي ﭘﯿﺶ ﺑﯿﻨﯽ ﻣﻘﺎدﯾﺮ ﺟﺮﯾﺎن روزاﻧﻪ در زﻣﺎن 1+t در ﯾﮏ دوره 23 ﺳﺎﻟﻪ اﺳﺘﻔﺎده و از 75 درﺻﺪ دادهﻫﺎ ﺑﻪﻣﻨﻈﻮر آﻣﻮزش ﻣﺪلﻫﺎ و از 25 درﺻﺪ ﺑﺎﻗﯽ ﻣﺎﻧﺪه ﺑﺮاي ﻣﺮﺣﻠﻪ آزﻣﻮن اﺳﺘﻔﺎده ﺷﺪ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد ﮐﻪ اﻟﮕﻮي ﺑﺮﺗﺮ در ﻫﺮ دو روش، ﻣﺪل ﺑﺎ ﻣﻘﺎدﯾﺮ ورودي ﺗﺎ ﺳﻪ ﮔﺎم زﻣﺎﻧﯽ ﺗﺄﺧﯿﺮ ﻣﯽﺑﺎﺷﺪ. ﻫﻤﭽﻨﯿﻦ، ﺑﺮ اﺳﺎس ﺳﻪ ﺷﺎﺧﺺ ارزﯾﺎﺑﯽ ﺿﺮﯾﺐ ﻫﻤﺒﺴﺘﮕﯽ )R(، ﻣﺠﺬور ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺧﻄﺎ )RMSE( و ﺿﺮﯾﺐ ﻧﺶ-ﺳﺎﺗﮑﻠﯿﻒ )E( در ﻣﺮﺣﻠﻪ آزﻣﻮن، روش ﺑﺮﻧﺎﻣﻪرﯾﺰي ﺑﯿﺎن ژن ﺑﺎ آﻣﺎرهﻫﺎي ارزﯾﺎﺑﯽ 0/902=RMSE=2/71 (m3s-1) ، R و 0/812=E ﻧﺴﺒﺖ ﺑﻪ روش ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ ﺑﺎ آﻣﺎرهﻫﺎي ارزﯾﺎﺑﯽ 0/905=m3s-1) ،R( 2/679=RMSE و 0/817=E داراي دﻗﺖ ﺑﺎﻻﺗﺮي ﻣﯽﺑﺎﺷﺪ. در ﺣﺎﻟﺖ ﮐﻠﯽ، ﻫﺮ دو روش داراي دﻗﺖ ﻗﺎﺑﻞ ﻗﺒﻮل و ﻧﺴﺒﺘﺎً ﯾﮑﺴﺎن ﻫﺴﺘﻨﺪ، وﻟﯽ ﺑﻪ دﻟﯿﻞ ﻣﺪلﺳﺎزي آﺳﺎنﺗﺮ روش ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ اﯾﻦ ﻣﺪل ﻣﯽﺗﻮاﻧﺪ ﺑﻪﻋﻨﻮان ﯾﮏ روش ﮐﺎرآﻣﺪ در ﭘﯿﺶﺑﯿﻨﯽ ﺟﺮﯾﺎن رودﺧﺎﻧﻪﻫﺎ ﻣﻮرد اﺳﺘﻔﺎده ﻗﺮار ﮔﯿﺮد.

The correct and accurate estimation of river flow can play an important role in reducing the effects of flood damage. In this research, Gene Expression Programming (GEP) model and Bayesian Network (BN) were used to predict daily flow of Mahabad River in Urmia Lake Basin. Accordingly, four input models with a delay of one to four days used to estimate daily flow at time t+1 over a 23-years period and 75% of data was used to train the models and 25% of the remaining data was used for the test stage. Results showed that the best model in both methods was the input pattern with three-time lags. Also, based on the correlation coefficient (R), Root Mean Square Error (RMSE) and Nash-Sutcliffe (E) coefficient in the test stage of the GEP method with R=0.902, RMSE=2.71(m3s-1) and E=0.812 compared to the BN method with R=0.905, RMSE=2.679(m3s-1( and E=0.817 is more accurate. In general, both methods have acceptable accuracy and are they relatively similar, but because of the simpler modeling, Bayesian Network method can be used as an efficient method for predicting river flow.