مقاﯾﺴﻪ و ارزﯾﺎﺑﯽ ﮐﺎراﺋﯽ ﻣﺪلﻫﺎي داده ﻣﺒﻨﺎ ﺟﻬﺖ ﺗﺨﻤﯿﻦ رﺳﻮب ﻣﻌﻠﻖ ﭘﺎﯾﯿﻦ دﺳﺖ ﺳﺪ درودزن

Comparison and evaluation of the performance of data-driven models for estimating suspended sediment downstream of Doroodzan Dam

ﺳﺪﻫﺎ ﺑﺮ ﺣﺴﺐ اﺑﻌﺎد ﺧﻮد ﺑﺎ اﯾﺠﺎد ﻣﺤﯿﻂﻫﺎي ﺳﺎﮐﻦ ﺑﺨﺶ اﻋﻈﻤﯽ از رﺳﻮب ورودي ﺑﻪ ﻣﺨﺰن را ﻣﻬﺎر ﻣﯽﮐﻨﻨﺪ. ﺑﺎ اﯾﻦ وﺟﻮد رﺳﻮب ﺧﺮوﺟﯽ از ﺳﺪ ﺑﻪ ﻋﻮاﻣﻞ ﻣﺨﺘﻠﻔﯽ ﻣﺎﻧﻨﺪ روش ﻣﺪﯾﺮﯾﺖ ﺳﺪ، رﺳﻮب ورودي، ارﺗﻔﺎع آب در ﻣﺨﺰن ﺷﮑﻞ ﻣﺨﺰن و دﺑﯽ ﺗﺨﻠﯿﻪ ﺑﺴﺘﮕﯽ دارد. در اﯾﻦ ﺗﺤﻘﯿﻖ ﻣﯿﺰان رﺳﻮب ﻣﻌﻠﻖ ﺧﺮوﺟﯽ از ﺳﺪ درودزن ﺑﺮ اﺳﺎس دوره آﻣﺎري 25 ﺳﺎﻟﻪ ﺑﺎ اﺳﺘﻔﺎده از ﺳﻪ روش ﯾﺎدﮔﯿﺮي ﺑﺮ اﺳﺎس اﻟﮕﻮرﯾﺘﻢ داده ﻣﺒﻨﺎ ﯾﻌﻨﯽ ﻧﺰدﯾﮑﺘﺮﯾﻦ ﻧﻘﺎط ﻫﻤﺴﺎﯾﻪ، رﮔﺮﺳﯿﻮنﮔﯿﺮي ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﻮرد ﺑﺮرﺳﯽ ﻗﺮار ﮔﺮﻓﺘﻪ اﺳﺖ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن ﻣﯽدﻫﺪ ﮐﻪ در روش ﻧﺰدﯾﮏﺗﺮﯾﻦ ﻫﻤﺴﺎﯾﻪ ﺗﻌﺪاد ﻫﻤﺴﺎﯾﻪ و وزن ﺮ ﭘﺎراﻣﺘﺮ ﺑﺮ دﻗﺖ ﻧﺘﺎﯾﺞ اﺛﺮ ﮔﺬار اﺳﺖ ﺑﻄﻮرﯾﮑﻪ در ﺑﯿﻦ ﺳﺎﺧﺘﺎرﻫﺎي ﻣﺨﺘﻠﻒ روش ﻧﺰدﯾﮑﺘﺮﯾﻦ ﻫﻤﺴﺎﯾﻪ، روش اﻧﺘﺨﺎب 6 ﺴﺎﯾﻪ ﺑﺎ اﻧﺘﺨﺎب وزنﻫﺎي 0/271 و 0/271 و 0/458 ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺮاي دﺑﯽ ورودي و دﺑﯽ رﺳﻮب ورودي و دﺑﯽ ﺧﺮوﺟﯽ ﻧﺘﺎﯾﺞ ﻣﻨﺎﺳﺒﺘﺮي را ﻧﺴﺒﺖ ﺑﻪ دﯾﮕﺮ ﺳﺎﺧﺘﺎرﻫﺎي اﯾﻦ روش ﻧﺸﺎن ﻣﯽدﻫﺪ. در ﺑﯿﻦ ﺳﺎﺧﺘﺎرﻫﺎي ﻣﺨﺘﻠﻒ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﺳﺎﺧﺘﺎر ﺑﺎ 2 ﻻﯾﻪ ﻣﺨﻔﯽ و ﺗﻌﺪاد 4 و 7 ﮔﺮه ﺑﺘﺮﺗﯿﺐ در ﻻﯾﻪ ﻫﺎي اول و دوم دﻗﺖ ﺑﺎﻻﺗﺮي ﻧﺴﺒﺖ ﺑﻪ دﯾﮕﺮ ﺳﺎﺧﺘﺎرﻫﺎ ﻧﺸﺎن ﻣﯽدﻫﻨﺪ. ﻣﻘﺎﯾﺴﻪ ﻫﺮ ﺳﻪ روش ﻧﺸﺎندﻫﻨﺪه دﻗﺖ ﺑﺎﻻﺗﺮ روش ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻧﺴﺒﺖ ﺑﻪ دو روش دﯾﮕﺮ اﺳﺖ.

Dams control most of the sediment entering the reservoir by creating static environments. However, sediment leaving the dam depends on various factors such as dam management method, inlet sediment, water height in the reservoir, the shape of the reservoir, and discharge flow. In this research, the amount of suspended sediment of Doroodzan Dam based on a statistical period of 25 years has been investigated using three learning methods based on the data-driven algorithm, namely the K nearest neighbors, regression, and neural network. The results show that among different structures of the K nearest neighbors, the selection of 6 neighborhoods has more precise outcomes than other structures. Also, among different structures of neural networks, a structure with two hidden layers and 4 and 7 nodes in each hidden layer respectively, predicted suspended sediment more accurately than other neural network structures. Comparison of different algorisms was indicated that neural networks have more accurate results than other mentioned methods.