ارزيابي رتبه‌اي دو رويكرد مدل‌سازي داده‌مبناء و مفهومي فرآيند بارش- رواناب در مقياس زماني ماهانه

Ranking Evaluation of Data-driven and Conceptual Modelling of Rainfall-Runoff Process in Monthly Time Scale

ﻣﺪلﺳﺎزي ﻣﺎﻫﺎﻧﻪ ﻓﺮآﯾﻨﺪ ﺑﺎرش- رواﻧﺎب ﻧﻘﺶ ﻣﻬﻤﯽ در ﺑﻬﺮهﺑﺮداري از ﺳﺪﻫﺎ دارد. در ﻣﻘﺎﻟﻪ ﺣﺎﺿﺮ ﮐﺎراﯾﯽ ﺳﻪ ﻣﺪل دادهﻣﺒﻨﺎء ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ )ANN(، ﺷﺒﮑﻪ ﻋﺼﺒﯽ رﮔﺮﺳﯿﻮن ﺗﻌﻤﯿﻢﯾﺎﻓﺘﻪ )GRNN( و K ﻧﺰدﯾﮏﺗﺮﯾﻦ ﻫﻤﺴﺎﯾﮕﯽ )KNN( در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﻣﺪل ﻣﻔﻬﻮﻣﯽ IHACRES در ﻣﺪلﺳﺎزي ﻣﺎﻫﺎﻧﻪ ﺑﺎرش- رواﻧﺎب ﺑﺎ دادهﻫﺎي ﻣﺸﺎﺑﻪ و ﺳﺎﺧﺘﺎر ﺑﻬﯿﻨﻪ ﻣﻮرد ارزﯾﺎﺑﯽ ﻗﺮار ﮔﺮﻓﺖ. ﺷﺒﯿﻪﺳﺎزي ﺟﺮﯾﺎن ﻣﺎﻫﺎﻧﻪ ورودي ﺑﻪ ﺳﺪ ﮐﺮﺧﻪ ﺑﻪ ﻋﻨﻮان ﻣﻄﺎﻟﻌﻪ ﻣﻮردي اﻧﺘﺨﺎب و از دادهﻫﺎي ﻣﺸﺎﻫﺪهاي 32 ﺳﺎﻟﻪ )1393-1361( دﻣﺎ و ﺑﺎرش ﻣﺎﻫﺎﻧﻪ و ﺟﺮﯾﺎن ﻣﺎﻫﺎﻧﻪ ورودي ﺑﻪ ﺳﺪ اﺳﺘﻔﺎده ﺷﺪ. ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻣﺘﻔﺎوت ﺑﻮدن اﻟﮕﻮﻫﺎي ﺑﺎرش-رواﻧﺎب در ﻣﺎهﻫﺎي ﻣﺨﺘﻠﻒ، دو ﻧﻮع ارزﯾﺎﺑﯽ ﮐﻠﯽ و ﻣﺎﻫﺎﻧﻪ از ﮐﺎراﯾﯽ ﻣﺪلﻫﺎ ﺑﺎ اﺳﺘﻔﺎده از روش رﺗﺒﻪدﻫﯽ و ﺑﺮ ﻣﺒﻨﺎي ﺳﻪ ﺷﺎﺧﺺ ارزﯾﺎﺑﯽ ﻧﺶ- ﺳﺎﺗﮑﻠﯿﻒ )NSE(، ﺟﺬر ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺧﻄﺎ )RMSE( و ﺿﺮﯾﺐ ﻫﻤﺒﺴﺘﮕﯽ )R( اﻧﺠﺎم ﺷﺪ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد ﮐﻪ از ﻫﺮ دو روش ارزﯾﺎﺑﯽ ﻣﺪلﻫﺎ در ﻣﺮﺣﻠﻪ ﺻﺤﺖﺳﻨﺠﯽ، دو ﻣﺪل ANN و KNN ﺑﻪ ﺗﺮﺗﯿﺐ داراي ﺑﯿﺸﺘﺮﯾﻦ و ﮐﻢﺗﺮﯾﻦ ﮐﺎراﯾﯽ در ﺗﺨﻤﯿﻦ ﺟﺮﯾﺎن ﻣﺎﻫﺎﻧﻪ ﺑﻮدﻧﺪ. ﺑﺮ اﺳﺎس ارزﯾﺎﺑﯽ ﮐﻠﯽ رﺗﺒﻪاي ﻣﺪلﻫﺎ، ﮐﺎراﯾﯽ دو ﻣﺪل NSE=0/749)ANN و 0/868=R( و NSE=0/699) IHACRES و 0/842=R( ﺑﺎ ﮐﺴﺐ 8 اﻣﺘﯿﺎز ﻣﺸﺎﺑﻪ ﺑﻮد و دو ﻣﺪل NSE=0/618) GRNN و 0/793=R( و NSE=0/601) KNN و 0/777=R( ﺑﺎ ﮐﺎراﯾﯽ ﻣﺸﺎﺑﻪ )5 اﻣﺘﯿﺎز( در رﺗﺒﻪ دوم ﻗﺮار ﮔﺮﻓﺘﻨﺪ. در ﺣﺎﻟﯿﮑﻪ ﺑﺮ اﺳﺎس روش ارزﯾﺎﺑﯽ رﺗﺒﻪاي ﻣﺎﻫﺎﻧﻪ، دو ﻣﺪل IHACRES و GRNN ﺑﺎ ﮐﺴﺐ ﻣﺠﻤﻮع 38 اﻣﺘﯿﺎز ﻣﺴﺎوي از ﺳﻪ ﺷﺎﺧﺺ ارزﯾﺎﺑﯽ ﺧﻄﺎ داراي ﮐﺎراﯾﯽ ﻣﺸﺎﺑﻪ ﺑﻮده و ﮐﺎراﯾﯽ آﻧﻬﺎ ﭘﺲ از ﻣﺪل ANN ﺑﺎ 48 اﻣﺘﯿﺎز در ﻣﻘﺎم دوم ﻗﺮار ﮔﺮﻓﺖ.

Rainfall-runoff monthly modelling process plays an important role in dams’ operation. Herein the performances of three data-based models including Artificial Neural Network (ANN), Generalized Regression Neural Network (GRNN) and K-Nearest Neighbor (KNN) are compared in tandem with IHACRES conceptual model, while they were applied with similar data, and optimal structures. Simulation of monthly inflow to Karkheh reservoir, Iran, was considered as the case study, and 32-year data (1982-2014) of monthly temperature and precipitation belong to the upper sub-basin of the dam, and monthly inflow to the reservoir were used. With respect to the different rainfall-runoff patterns in different months, the models assessed in a general and monthly manners using a rating method based on performance criteria including: Nash-Sutcliff Efficiency (NSE), Root Mean Square Error (RMSE) and Correlation Coefficient(R). Results showed that both model evaluation procedure in validation phase, ANN and KNN models have the highest and lowest efficiency in monthly streamflow forecasting, respectively. Based on the rating general evaluation the performance of ANN (NSE= 0.749, R= 0.868) and IHACRES (NSE= 0.699, R= 0.842) are similar with a score of 8 while the GRNN (NSE= 0.618, R= 0.793) and KNN (NSE= 0.601, R= 0.777) models with similar performance (score 5) were ranked in the second order. However, in accordance with rating monthly assessment of the models, the performance of GRNN was similar to IHACRES with the total score of 38 based on three criteria while they were ranked in the second order after ANN model with score 48.