ارزيابي قابليت روش تجزيه متعامد سره جهت تعيين ورودي به مدل شبكه عصبي براي پيش‌بيني جريان ماهانه ورودي به سد علويان

Proper Orthogonal Decomposition Performance to Determine the Inputs to the Artificial Neural Network for Prediction of Inflow into Alavian Dam

زمينه و ﻫﺪف: ﺳﺪﻫﺎ ﺑﻪ ﻋﻨﻮان ﯾﮑﯽ از اﺳﺎﺳﯽ ﺗﺮﯾﻦ ﻣﻨﺎﺑﻊ ﺗﺎﻣﯿﻦ آب ﺷﺮب، ﮐﺸﺎورزي، ﺑﺮق آﺑﯽ و ﺻﻨﻌﺘﯽ از ﻧﻘﺶ ﻣﻬﻤﯽ در ﺗﻮﺳﻌﻪ ﺟﻮاﻣﻊ اﻧﺴﺎﻧﯽ و ﻣﺤﯿﻂزﯾﺴﺖ اﻃﺮاف ﺧﻮد ﺑﺮﺧﻮردارﻧﺪ. ﺑﻨﺎﺑﺮاﯾﻦ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻧﻘﺶ اﺳﺎﺳﯽ ﺳﺪﻫﺎ در ﭘﻮﯾﺎﯾﯽ ﻣﺤﯿﻂ اﻃﺮاف ﺧﻮد، ﺑﺮآورد ﺟﺮﯾﺎن ورودي ﺑﻪ آن ﻫﺎ از اﻫﻤﯿﺖ وﯾﮋه اي ﺑﺮﺧﻮردار ﺑﻮده و از اﺑﺰارﻫﺎي ﻣﻬﻢ و ﻣﺆﺛﺮ در ﻣﺪﯾﺮﯾﺖ ﺑﻬﯿﻨﻪ ﮐﻤﯽ و ﮐﯿﻔﯽ ﻣﻨﺎﺑﻊ آب اﺳﺖ. روش ﺑﺮرﺳﯽ: در اﯾﻦ ﺗﺤﻘﯿﻖ ﺳﻌﯽ ﺷﺪه ﺗﺎ ﺑﺎ اﺳﺘﻔﺎده از ﻣﺪل ﻫﻮش ﻣﺼﻨﻮﻋﯽ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ )ANN( اﻗﺪام ﺑﻪ ﻣﺪل ﺳﺎزي ﺟﺮﯾﺎن ﻣﺎﻫﺎﻧﻪ رودﺧﺎﻧﻪ ﺻﻮﻓﯽ ﭼﺎي، ورودي ﺑﻪ ﺳﺪ ﻋﻠﻮﯾﺎن، ﮔﺮدد. ﻫﻤﭽﻨﯿﻦ ﺑﻪ ﻣﻨﻈﻮر ﺑﻬﺒﻮد ﻋﻤﻠﮑﺮد ﻣﺪل ﻣﺬﮐﻮر و ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ اﻃﻼﻋﺎت زﯾﺎد ورودي ﺑﻪ اﯾﻦ ﻣﺪل، از روش ﺗﺠﺰﯾﻪ ﻣﺘﻌﺎﻣﺪ ﺳﺮه )POD( ﺑﻪ ﻣﻨﻈﻮر ﺗﻌﯿﯿﻦ ﺑﻬﺘﺮﯾﻦ اﻟﮕﻮي ورودي ﺑﻪ ﻣﺪل ANN اﺳﺘﻔﺎده ﮔﺮدﯾﺪ. در ﻧﻬﺎﯾﺖ ﻧﯿﺰ ﻋﻤﻠﮑﺮد دو ﻣﺪل ANN و ﻣﺪل ﺗﺮﮐﯿﺒﯽ POD-ANN ﺑﺮ ﭘﺎﯾﻪ آﻣﺎره ﻫﺎي ﺿﺮﯾﺐ ﺗﻌﯿﯿﻦ )R (، ﻣﯿﺎﻧﮕﯿﻦ ﺧﻄﺎي ﻣﻄﻠﻖ )MAE( و ﻣﯿﺎﻧﮕﯿﻦ ﻗﺪرﻣﻄﻠﻖ ﺧﻄﺎي ﻧﺴﺒﯽ )AARE( ﻣﻮرد ارزﯾﺎﺑﯽ ﻗﺮار ﮔﺮﻓﺖ. ﯾﺎﻓﺘﻪ ﻫﺎ: ﻧﺘﺎﯾﺞ اﯾﻦ ﺗﺤﻘﯿﻖ ﻣﺸﺨﺺ ﻧﻤﻮد ﮐﻪ اﮔﺮﭼﻪ ﻣﻘﺎدﯾﺮ ﭘﯿﺶ ﺑﯿﻨﯽ ﺷﺪه دﺑﯽ ورودي ﺑﻪ ﻣﺨﺰن ﺳﺪ ﺗﻮﺳﻂ ﻣﺪل ANN ﻧﺰدﯾﮏ ﺑﻪ ﻣﻘﺎدﯾﺮ ﻣﺸﺎﻫﺪه اي ﻫﺴﺘﻨﺪ اﻣﺎ ﻋﻤﻠﮑﺮد آن در ﻧﻘﺎط ﺑﺎ دﺑﯽ ﺑﺎﻻ ﺑﺎ ﺧﻄﺎي ﻗﺎﺑﻞ ﺗﻮﺟﻬﯽ ﻫﻤﺮاه اﺳﺖ. ﻫﻤﭽﻨﯿﻦ ﯾﺎﻓﺘﻪ ﻫﺎي اﯾﻦ ﺗﺤﻘﯿﻖ ﺣﺎﮐﯽ از ﻋﻤﻠﮑﺮد ﺑﻬﺘﺮ ﻣﺪل POD-ANN ﻧﺴﺒﺖ ﺑﻪ ﻣﺪل ANN در ﻧﻘﺎط ﺑﺎ دﺑﯽ ﺑﺎﻻ ﺑﻮد. در ﺣﺎﻟﺖ ﮐﻠﯽ ﻧﺘﺎﯾﺞ ﺑﻪ دﺳﺖ آﻣﺪه از ﻣﺪل POD-ANN اﺟﺮا2ﺷﺪه ﻣﺸﺨﺺ ﻧﻤﻮد ﮐﻪ ﻣﻘﺪار آﻣﺎره ﻫﺎي MAE ،R و AARE ﻣﺪل در ﻫﺮ دو ﻣﺮﺣﻠﻪ واﺳﻨﺠﯽ و ﺻﺤﺖ ﺳﻨﺠﯽ ﺑﻬﺒﻮد ﻗﺎﺑﻞ ﺗﻮﺟﻬﯽ ﻧﺴﺒﺖ 2 ﺑﻪ ﻣﻘﺎدﯾﺮ ﻣﺸﺎﺑﻪ در ﻣﺪل ANN داﺷﺘﻪ اﻧﺪ. ﻣﻘﺪار آﻣﺎره ﻫﺎي MAE ،R و AARE در ﻣﺮﺣﻠﻪ ﺻﺤﺖ ﺳﻨﺠﯽ POD-ANN ﺑﻪ ﺗﺮﺗﯿﺐ ﻣﻌﺎدل 0/79 ،0/93 و 0/54 ﺑﻮد. ﺑﺤﺚ و ﻧﺘﯿﺠﻪ ﮔﯿﺮي: ﻋﻤﻠﮑﺮد ﺑﻬﺘﺮ ﻣﺪل POD-ANN در دﺑﯽ ﺑﺎ ﻣﻘﺎدﯾﺮ ﺑﺎﻻ ﻧﺴﺒﺖ ﺑﻪ ﻣﺪل ANN ﻣﯽ ﺗﻮاﻧﺪ ﺑﻪ دﻟﯿﻞ ﻋﻤﻞ ﭘﯿﺶ ﭘﺮدازش ﺑﺮ روي ﻣﺘﻐﯿﺮﻫﺎي ورودي و ﮐﺎﻫﺶ ﺗﻌﺪاد آن ﻫﺎ در ﻣﺪل POD-ANN در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﻣﺪل ANN ﺑﺎﺷﺪ. ﺑﻨﺎﺑﺮاﯾﻦ ﻣﯽ ﺗﻮان ﻧﺘﯿﺠﻪ ﮔﯿﺮي ﻧﻤﻮد ﮐﻪ ﻋﻤﻞ ﭘﯿﺶ ﭘﺮدازش ﺑﺮ روي ﻣﺘﻐﯿﺮﻫﺎي ورودي ﺑﻪ ﻣﺪل ANN و ﮐﺎﻫﺶ ﺗﻌﺪاد ﻣﺘﻐﯿﺮﻫﺎي ورودي ﺑﻪ اﯾﻦ ﻣﺪل ﻫﻤﺮاه ﺑﺎ ﺑﻬﺒﻮد ﻋﻤﻠﮑﺮد آن بوده است

Abstract Background and Objective: Dams play an important role in development of countries by drinking and agricultural water supply, flood control, hydropower energy supply and recreational purposes. Constructing a dam and making an artificial lake has an important effect on surrounding environment, so being able to forecast the inflow to the dam is an important issue for water resource management. Method: In this study artificial neural network (ANN) was applied to forecast the monthly inflow from Soofichai River to Alavian Dam. Regarding the huge amount of input data to ANN model and for optimizing its application, proper orthogonal decomposition (POD) was used in order to determine the best inputs for ANN model . Finally, the application of ANN and POD-ANN models was evaluated by determination coefficient (R2), mean absolute error (MAE) and average of absolute relative error (AARE). Findings: Results of ANN and POD-ANN models indicated that although ANN output is close to the observed values of inflow to the dam, but it has significant errors. POD-ANN model showed better results than ANN model for high values of inflow. In generall, comparing R2, MAE and AARE values of two models revealed that POD-ANN model had better performance in both calibration and verification steps in comparison with ANN model. R2, MAE and AARE in verification step of POD-ANN model were 0.93, 0.79, and 0.54, respectively. Discussion and Conclusion: Preprocessing data contributes to better performance of POD-ANN than ANN model, especially in high values of inflow. Therefore, it can be concluded that applying data preprocessing and reducing inputs to ANN model enhances its performance.