‫ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﺩﺍﺩﻩﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺑﻪ ﺷﺒﮑﻪ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻲ‬

‫ﺑﻪﻣﻨﻈﻮﺭ ﺑﺮﺁﻭﺭﺩ ﻣﻘﺪﺍﺭ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ‬

۲ ۱
‫ﻣﻴﻨﺎ ﺳﻴﺎﺡﻓﺮﺩ‬ ‫ﻭﺣﻴﺪ ﻧﻮﺭﺍﻧﻲ‬

(۹۰/۱۲/۲۰‫ﭘﺬﻳﺮﺵ‬ ۹۰/۱۰/۲۵ ‫ﺁﺧﺮﻳﻦ ﺍﺻﻼﺣﺎﺕ ﺩﺭﻳﺎﻓﺘﻲ‬ ۸۹/۱۰/۱ ‫)ﺩﺭﻳﺎﻓﺖ‬

‫ﭼﻜﻴﺪﻩ‬
‫ ﺍﻳﻦ ﻣﻘﺎﻟﻪ ﮐﺎﺭﺑﺮﺩ ﺷﺒﮑﻪﻫﺎﻱ ﻋﺼﺒﻲ‬.‫ﺑﺮﺁﻭﺭﺩ ﻣﻘﺎﺩﻳﺮ ﺗﺒﺨﻴﺮ ﺑﺮﺍﻱ ﻣﺪﻳﺮﻳﺖ ﮐﺎﺭﺁﻣﺪ ﻣﻨﺎﺑﻊ ﺁﺏ ﺩﺭ ﻣﻨﺎﻃﻖ ﺑﺎ ﺍﻗﻠﻴﻢ ﻧﻴﻤﻪ ﺧﺸﮏ ﺍﻣﺮﻱ ﺿﺮﻭﺭﻱ ﺍﺳﺖ‬
‫ ﺁﻳﺮﻭﺩﻳﻨﺎﻣﻴﮑﻲ ﻭ ﺗﺮﮐﻴﺒﻲ ﭘﻨﻤﻦ ﺭﺍ ﺑﻪﻣﻨﻈـﻮﺭ ﺑـﺮﺁﻭﺭﺩ ﻣﻘـﺎﺩﻳﺮ ﺗﺒﺨﻴـﺮ‬،‫ ﺭﮔﺮﺳﻴﻮﻥ ﺧﻄﻲ ﭼﻨﺪ ﻣﺘﻐﻴﺮﻩ ﻭ ﺳﻪ ﺭﻭﺵ ﺗﺠﺮﺑﻲ ﺗﻌﺎﺩﻝ ﺍﻧﺮﮊﻱ‬،‫ﻣﺼﻨﻮﻋﻲ‬
‫ ﻋﻼﻭﻩ ﺑﺮ ﺍﻳﻦ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﺑﻪ ﺩﻭ ﺭﻭﺵ ﻣﺸـﺘﻘﺎﺕ ﺟﺰﺋـﻲ ﻭ ﻭﺯﻥﻫـﺎﻱ‬.‫ﺭﻭﺯﺍﻧﻪ ﺗﺸﺘﮏ ﺑﺮﺍﻱ ﺩﻭ ﺷﻬﺮ ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ ﻣﻮﺭﺩ ﺍﺭﺯﻳﺎﺑﻲ ﻗﺮﺍﺭ ﺩﺍﺩ‬
‫ﺍﺗﺼﺎﻟﻲ ﺑﺮﺍﻱ ﺗﺒﻴﻴﻦ ﻣﻴﺰﺍﻥ ﺍﺛﺮ ﻫﺮ ﭘﺎﺭﺍﻣﺘﺮ ﻭﺭﻭﺩﻱ ﺍﺯ ﻧﻈﺮ ﻣﻘﺪﺍﺭ ﻭ ﺟﻬﺖ ﺭﻭﻱ ﻧﺘﺎﻳﺞ ﻣﺪﻝﻫﺎﻱ ﺷﺒﮑﻪ ﻋﺼﺒﻲ ﻭ ﻧﻴﺰ ﺗﺸﺨﻴﺺ ﺑﻬﺘـﺮﻳﻦ ﺗﺮﮐﻴﺒـﺎﺕ‬
،‫ ﺗﺸﻌﺸـﻊ ﺧﻮﺭﺷـﻴﺪ‬،‫ ﺗﺒﺨﻴﺮ ﺗﺸـﺘﮏ‬،‫ﻴﺘﻬﺎﻱ ﺩﻣﺎﻱ ﻫﻮﺍ‬‫ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻫﻴﺪﺭﻭﻟﻮﮊﻳﮑﻲ ﺍﺳﺘﻔﺎﺩﻩ ﺷﺪﻩ ﺷﺎﻣﻞ ﻣﺸﺎﻫﺪﺍﺕ ﺭﻭﺯﺍﻧﻪ ﮐﻤ‬.‫ﻭﺭﻭﺩﻱ ﺍﻧﺠﺎﻡ ﺷﺪ‬
.‫ ﻧﺘﺎﻳﺞ ﺭﻭﺷﻬﺎﻱ ﮐﻼﺳﻴﮏ ﻭ ﻣﺪﻝﻫﺎﻱ ﺷﺒﮑﻪ ﻋﺼﺒﻲ ﺑـﺎ ﻣﻘـﺎﺩﻳﺮ ﺗﺒﺨﻴـﺮ ﺭﻭﺯﺍﻧـﻪ ﻣﻘﺎﻳﺴـﻪ ﺷـﺪ‬.‫ ﺭﻃﻮﺑﺖ ﻧﺴﺒﻲ ﻭ ﺳﺮﻋﺖ ﺑﺎﺩ ﺑﻮﺩﻧﺪ‬،‫ﻓﺸﺎﺭ ﻫﻮﺍ‬
‫ ﻧﺘﺎﻳﺞ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﻧﺸـﺎﻥ ﺩﺍﺩ‬.‫ﻣﻘﺎﻳﺴﻪﻫﺎ ﺑﻬﺘﺮ ﺑﻮﺩﻥ ﭘﺎﺳﺦ ﺷﺒﮑﻪﻫﺎﻱ ﻋﺼﺒﻲ ﺩﺭ ﺑﺮﺁﻭﺭﺩ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﻧﺴﺒﺖ ﺑﻪ ﺳﺎﻳﺮ ﻣﺪﻝﻫﺎ ﺭﺍ ﻧﺸﺎﻥ ﺩﺍﺩ‬
‫ ﺗﺸﻌﺸﻊ ﺧﻮﺭﺷﻴﺪﻱ ﻭ ﺗﺒﺨﻴﺮ ﺭﻭﺯ ﮔﺬﺷﺘﻪ ﺑﻴﺸﺘﺮﻳﻦ ﺗﺄﺛﻴﺮ ﺭﺍ ﺭﻭﻱ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﻧﺴﺒﺖ ﺑﻪ ﺳﺎﻳﺮ ﻋﻮﺍﻣﻞ ﺩﺭ ﻫﺮ ﺩﻭ ﻣﻨﻄﻘـﻪ‬،‫ﮐﻪ ﺩﺭﺟﻪ ﺣﺮﺍﺭﺕ ﻫﻮﺍ‬
.‫ﺩﺍﺭﻧﺪ ﻭ ﺗﺄﺛﻴﺮ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺩﻳﮕﺮ ﮐﻢ ﺍﺳﺖ‬

‫ ﺍﺭﻭﻣﻴﻪ‬،‫ ﺗﺒﺮﻳﺰ‬،‫ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ‬،‫ ﺷﺒﮑﻪﻫﺎﻱ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻲ‬،‫ ﺁﻧﺎﻟﻴﺰﺣﺴﺎﺳﻴﺖ‬:‫ﻭﺍﮊﻩﻫﺎﻱ ﮐﻠﻴﺪﻱ‬

Sensitivity Analysis of ANN Inputs in Estimating Daily Evaporation

Vahid Nourani 1 Mina Sayyah Fard 2

(Received Dec. 22, 2010 Revised Jan. 15, 2012 Accepted March 10, 2012)

Abstract
Estimation of evaporation values is needed for efficient management of water resources at semi-arid regions.
This paper presents application of Artificial Neural Networks (ANNs), Multiple Linear Regression (MLR) and
empirical models viz.: Energy balance ، Aerodynamic ، Penman for estimation of daily pan evaporation for
Tabriz and Urmia cities. Furthermore, in order to determine the effect of each input parameter on the output
variable in terms of magnitude and direction and also identify the best combinations of the model inputs, two
sensitivity analysis methods i.e. the Partial Derivation method (PaD) and the Weights method have been applied
on the ANNs results. The used hydrological variables include daily observations of air temperature, pan
evaporation, solar radiation, air pressure, relative humidity, and wind speed. The results of the classic methods
and ANN models are compared to daily observations of evaporation values. The comparison showed that there is
better agreement between the ANN estimations and measurements of daily evaporation than other models.
Sensitivity analysis results showed that air temperature, solar radiation and previous day evaporation have
maximum effects on daily evaporation in both regions and the contributions of the other variables are
insignificant.

Keywords: Sensitivity Analysis, Artificial Neural Networks, Daily Evaporation, Tabriz, Urmia.
1. Assoc. Prof. of Water Eng., Dept. of Civil Eng., Tabriz University, Tabriz ‫ ﺩﺍﻧﺸﮕﺎﻩ ﺗﺒﺮﻳﺰ )ﻧﻮﻳﺴﻨﺪﻩ‬،‫ ﺩﺍﻧﺸﻜﺪﻩ ﻣﻬﻨﺪﺳﻲ ﻋﻤﺮﺍﻥ‬،‫ ﺩﺍﻧﺸﻴﺎﺭ ﮔﺮﻭﻩ ﻣﻬﻨﺪﺳﻲ ﻋﻤﺮﺍﻥ ﺁﺏ‬- ١
(Corresponding Author) (+98 411) 3392409 vnourani@umn.edu vnourani@umn.edu ( ٠٤١١) ٣٣٩٢٤٠٩ (‫ﻣﺴﺌﻮﻝ‬
2. M.Sc. Student of Hydraulic Structures, Dept. of Civil Eng., Tabriz
University, Tabriz ‫ ﺩﺍﻧﺸﮕﺎﻩ‬،‫ ﺩﺍﻧﺸﻜﺪﻩ ﻣﻬﻨﺪﺳﻲ ﻋﻤﺮﺍﻥ‬،‫ ﺩﺍﻧﺸﺠﻮﻱ ﮐﺎﺭﺷﻨﺎﺳﻲ ﺍﺭﺷﺪ ﺳﺎﺯﻩ ﻫﺎﻱ ﻫﻴﺪﺭﻭﻟﻴﮑﻲ‬- ٢
‫ﺗﺒﺮﻳﺰ‬

١٣٩٢ ‫ ﺳﺎﻝ‬٣ ‫ﺷﻤﺎﺭﻩ‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬  

 ‫ﺍﺯ ﻧﻈﺮ ﻣﻘﺪﺍﺭ ﻭ ﺟﻬﺖ ﺩﺭ ﻛﻞ ﻓﻀـﺎﻱ ﻭﺭﻭﺩﻱ ﻗﺎﺑـﻞ ﺗﻌﻴـﻴﻦ ﻧﻴﺴـﺖ‪ .‬ﺍﺯ‬
‫ﺍﻳﻦﺭﻭ ﺁﻧﻬﺎ ﺑﺎ ﺭﻭﺷﻲ ﻣﺒﺘﻨﻲ ﺑﺮ ﺁﻧـﺎﻟﻴﺰ ﺁﻣـﺎﺭﻱ ﻣﺸـﺘﻖ ﺧﺮﻭﺟـﻲ ﺷـﺒﻜﻪ‬
‫ﻧﺴﺒﺖ ﺑﻪ ﻭﺭﻭﺩﻱ )ﺣﺴﺎﺳﻴﺖ ﺧﺮﻭﺟﻲ ﺑﻪ ﻭﺭﻭﺩﻱ(‪ ،‬ﺁﻧـﺎﻟﻴﺰ ﺣﺴﺎﺳـﻴﺘﻲ‬
‫ﺑﺮ ﺭﻭﻱ ﻣﺴﺌﻠﻪ ﺑﻬﺮﻩﻭﺭﻱ ﺩﺭ ﺗﻮﻟﻴﺪ ﻟﻮﻟـﻪﻫـﺎﻱ ﺍﻧﺘﻘـﺎﻝ ﺍﻧﺠـﺎﻡ ﺩﺍﺩﻩﺍﻧـﺪ ﻭ‬
‫ﺑﻴﺎﻥ ﻛﺮﺩﻩﺍﻧﺪ ﻛﻪ ﻣﻄﺎﻟﻌﻪ ﺑﺮ ﺭﻭﻱ ﺭﻭﺍﺑﻂ ﺩﺭﻭﻥ ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ ﺑﺎﻋـﺚ‬
‫ﺁﺳﺎﻥﺗﺮ ﺷﺪﻥ ﺩﺭﮎ ﻭ ﺗﻔﺴﻴﺮ ﻧﺘﺎﻳﺞ ﻣﺪﻝﺳﺎﺯﻱ ﻣﻲﺷﻮﺩ]‪.[۴‬‬
‫ﺩﺭ ﺍﺭﺗﺒﺎﻁ ﺑﺎ ﺗﺤﻘﻴﻘـﺎﺕ ﺻـﻮﺭﺕ ﮔﺮﻓﺘـﻪ ﺩﺭ ﻣـﻮﺭﺩ ﺑـﺮﺁﻭﺭﺩ ﻣﻴـﺰﺍﻥ‬
‫ﺗﺒﺨﻴﺮ ﻭ ﺗﻌﺮﻕ ﺗﻮﺳﻂ ﻣﺪﻝﻫـﺎﻱ ‪ ANN‬ﻣـﻲ ﺗـﻮﺍﻥ ﺑـﻪ ﻣـﻮﺍﺭﺩ ﻣﺘﻨـﻮﻋﻲ‬
‫ﺍﺷﺎﺭﻩ ﮐﺮﺩ‪ .‬ﺩﺭ ﺳﺎﻝ ‪ ۲۰۰۸‬ﺗﻜﻨﻮﻟﻮﮊﻱ ‪ ANN‬ﺑـﺮﺍﻱ ﺗﺨﻤـﻴﻦ ﺗﺒﺨﻴـﺮ ﻭ‬
‫ﺗﻌـﺮﻕ ﺑـﺎ ﻣﺘﻐﻴﺮﻫـﺎﻱ ﻭﺭﻭﺩﻱ ﻣﺤــﺪﻭﺩ ﺑـﻪﻛـﺎﺭﺑﺮﺩﻩ ﺷـﺪﻩ ﺍﺳــﺖ]‪ .[۹‬ﺩﺭ‬
‫ﺗﺨﻤﻴﻦ ﺗﺒﺨﻴﺮ ﻭ ﺗﻌﺮﻕ ﻧﻴﺎﺯ ﺑﻪ ﻣﺘﻐﻴﺮﻫـﺎﻱ ﺍﻗﻠﻴﻤـﻲ ﺯﻳـﺎﺩﻱ ﺍﺳـﺖ ﺣـﺎﻝ‬
‫ﺁﻧﮑﻪ ﺗﺤﻘﻴﻖ ﺑﺎﻻ ﺩﺭ ﭘـﻲ ﺑﺮﺭﺳـﻲ ﺩﻗـﺖ ﻳـﮏ ﻣـﺪﻝ ‪ ANN‬ﺩﺭ ﺗﺨﻤـﻴﻦ‬
‫ﺗﺒﺨﻴﺮ ﻭ ﺗﻌﺮﻕ ﺑﺎ ﻓﻀـﺎﻱ ﻭﺭﻭﺩﻱ ﻣﺤـﺪﻭﺩ ﺑـﻮﺩﻩ ﺍﺳـﺖ‪ .‬ﺩﺭ ﺍﻗﻠـﻴﻢﻫـﺎﻱ‬
‫ﺧﺸﮏ ﻭ ﻧﻴﻤﻪﺧﺸﮏ ﻧﻘﺶ ﺗﺒﺨﻴﺮ ﺩﺭ ﺗﻮﺳـﻌﻪ ﻣﻨـﺎﺑﻊ ﺁﺑـﻲ ﭘـﺮﺭﻧـﮓﺗـﺮ‬
‫ﻣﻲﺷﻮﺩ‪ .‬ﻧﺘﺎﻳﺞ ﺣﺎﺻﻞ ﺍﺯ ﺑﻪﮐﺎﺭ ﺑﺮﺩﻥ ﺭﺍﺑﻄﻪﻫـﺎﻱ ﺗﺠﺮﺑـﻲ ﺑـﺎ ﺗﻮﺟـﻪ ﺑـﻪ‬
‫ﭘﻴﭽﻴﺪﻩ ﺑﻮﺩﻥ ﻓﺮﺍﻳﻨﺪ ﺗﺒﺨﻴﺮ ﻭ ﺩﺍﺩﻩﻫﺎﻱ ﺁﻣﺎﺭﻱ ﺯﻳﺎﺩ ﻣﻮﺭﺩ ﻧﻴـﺎﺯ‪ ،‬ﭼﻨـﺪﺍﻥ‬
‫ﺭﺿﺎﻳﺖﺑﺨﺶ ﻧﻴﺴﺖ‪ .‬ﻣﻘﺪﻡﻧﻴﺎ ﻭ ﻫﻤﮑﺎﺭﺍﻥ ﺩﺭ ﺳـﺎﻝ ‪ ۲۰۰۸‬ﻣـﺪﻝﻫـﺎﻱ‬
‫‪ ANN‬ﻭ ﺳﻴﺴﺘﻢ ﺍﺳﺘﻨﺘﺎﺝ ﻓﺎﺯﻱ ﺗﻄﺒﻴﻘﻲ‪ ٦‬ﺭﺍ ﺑﺮﺍﻱ ﺗﺨﻤﻴﻦ ﺗﺒﺨﻴﺮ ﺑﻪﮐـﺎﺭ‬
‫ﺑﺮﺩﻩﺍﻧﺪ ﮐـﻪ ﻧﺘـﺎﻳﺞ ﺑﺮﺗـﺮ ﺑـﻮﺩﻥ ﺍﻳـﻦ ﺩﻭ ﺗﮑﻨﻴـﮏ ﺭﺍ ﺩﺭ ﺗﺨﻤـﻴﻦ ﺗﺒﺨﻴـﺮ‬
‫ﻧﺴﺒﺖ ﺑﻪ ﺭﺍﺑﻄﻪﻫﺎﻱ ﺗﺠﺮﺑﻲ ﻧﺸﺎﻥ ﻣﻲﺩﻫﺪ]‪.[۱۰‬‬
‫ﺳــﻴﺎﻩ ‪ ،‬ﺍﺳــﺘﻔﺎﺩﻩ ﺍﺯ ﺷــﺒﮑﻪﻫــﺎﻱ ﻋﺼــﺒﻲ ﻣﺼــﻨﻮﻋﻲ ﺍﺳــﺖ‪ANNs .‬‬
‫ﺩﺭ ﺭﺍﺳﺘﺎﻱ ﻣﻄﺎﻟﻌﻪ ﺗﺄﺛﻴﺮ ﺗﺮﮐﻴﺒﺎﺕ ﻣﺨﺘﻠﻒ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻱ ﺍﻗﻠﻴﻤـﻲ ﺑـﺮ‬
‫ﺗﻠﻔﺎﺕ ﺗﺒﺨﻴﺮ ﻣﺨﺰﻥ ﺳﺪ‪ ،‬ﺩﻳﺴـﻮﺍﻝ‪ ٧‬ﻭ ﭘـﺎﻝ ‪ ٨‬ﺩﺭ ﺳـﺎﻝ ‪ ۲۰۰۸‬ﺗﮑﻨﻴـﮏ‬
‫ﻣــﺪﻝﺳــﺎﺯﻱ ‪ ANN‬ﺭﺍ ﺑــﻪﮐــﺎﺭﺑﺮﺩﻩﺍﻧــﺪ]‪ .[۱۱‬ﺩﺭ ﻣﺮﮐــﺰ ﺗﺤﻘﻴﻘــﺎﺕ‬
‫ﮐﺸﺎﻭﺭﺯﻱ ﺧﻮﺯﺳـﺘﺎﻥ ﻭ ﻧﻴـﺰ ﺩﺭ ﻣﻨﻄﻘـﻪﺍﻱ ﺩﺭ ﺗـﺎﻳﻮﺍﻥ‪ ،‬ﺍﺑـﺰﺍﺭ ‪ ANN‬ﻭ‬
‫ﻣﻌﺎﺩﻟﻪ ﺗﺠﺮﺑـﻲ ﻫـﺎﺭﮔﺮﻳﻮﺯ‪ ٩‬ﺑـﻪﻣﻨﻈـﻮﺭ ﺑـﺮﺁﻭﺭﺩ ﻣﻴـﺰﺍﻥ ﺗﺒﺨﻴـﺮ ﺭﻭﺯﺍﻧـﻪ‬
‫ﺗﺸﺘﮏ ﺑـﻪﮐـﺎﺭ ﮔﺮﻓﺘـﻪ ﺷـﺪﻩﺍﻧـﺪ]‪ ۱۲‬ﻭ ‪ .[۱۳‬ﻳـﮏ ﻣـﺪﻝ ‪ ANN‬ﺑـﺮﺍﻱ‬
‫ﺗﺨﻤﻴﻦ ﻣﻴﺰﺍﻥ ﺗﺒﺨﻴـﺮ ﺁﺏ ﻣﻮﺟـﻮﺩ ﺩﺭ ﺧـﺎﮎ ﻣﺤـﻴﻂ ﮔﻠﺨﺎﻧـﻪ ﺩﺭ ﭼـﻴﻦ‬
‫ﺑﻪﮐﺎﺭ ﮔﺮﻓﺘﻪ ﺷﺪﻩ ﺍﺳـﺖ ﮐـﻪ ﻧﺘـﺎﻳﺞ‪ ،‬ﺩﻗـﺖ ﺑـﺎﻻﻱ ﺍﻳـﻦ ﺭﻭﺵ ﺭﺍ ﻧﺸـﺎﻥ‬
‫ﻣﻲﺩﻫﺪ]‪ .[۱۴‬ﻃﺒﺮﻱ ﻭ ﻫﻤﻜﺎﺭﺍﻥ ﺩﺭ ﺗﺤﻘﻴﻖ ﺧﻮﺩ ﺑﺮﺍﻱ ﺗﺨﻤـﻴﻦ ﻣﻴـﺰﺍﻥ‬
‫ﺗﺒﺨﻴﺮ ﺍﺯ ﺗﺸﺘﮏ ﺩﺭ ﻧﺎﺣﻴﻪﺍﻱ ﺑﺎ ﺍﻗﻠﻴﻢ ﻧﻴﻤﻪ ﺧﺸﮏ ﺩﺭ ﺍﻳﺮﺍﻥ‪ ،‬ﻣـﺪﻝﻫـﺎﻱ‬
‫‪ ANN‬ﻭ ﺭﮔﺮﺳﻴﻮﻧﻲ ﺭﺍ ﺑﻪﮐﺎﺭ ﺑﺮﺩﻩﺍﻧﺪ ﮐﻪ ﻧﺘﺎﻳﺞ‪ ،‬ﺑﺮﺗﺮ ﺑﻮﺩﻥ ﻣـﺪﻝ ﺷـﺒﮑﻪ‬
‫ﺭﺍ ﺑــﻪ ﻣــﺪﻝ ﺭﮔﺮﺳــﻴﻮﻧﻲ ﺩﺭ ﺗﺨﻤــﻴﻦ ﻧﺰﺩﻳــﮏ ﺑــﻪ ﻭﺍﻗﻌﻴــﺖ ﻧﺸــﺎﻥ‬
‫ﻧﺮﻭﻥﻫﺎﻱ ﻣﻴﺎﻧﻲ ﻭ ﺧﺮﻭﺟﻲ‪ ،‬ﺍﺳـﺘﻔﺎﺩﻩ ﺍﺯ ﺁﻧـﺎﻟﻴﺰ ﺁﻣـﺎﺭﻱ ﻣﺆﻟﻔـﻪ ﺍﺻـﻠﻲ‬
‫ﻣﻲ ﺩﻫﺪ]‪ .[۱۵‬ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻱ ﺯﻳـﺎﺩﻱ ﺩﺭ ﺑـﺮﺁﻭﺭﺩ ﻣﻴـﺰﺍﻥ ﺗﺒﺨﻴـﺮ ﺩﺧﺎﻟـﺖ‬
‫ﻣﻲﻛﻨﻨﺪ ﻛﻪ ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﻣﺘﻔﺎﻭﺕ ﺑﻮﺩﻥ ﺷـﺪﺕ ﺗـﺄﺛﻴﺮ ﺁﻧﻬـﺎ ﺍﺯ ﺍﻗﻠﻴﻤـﻲ ﺑـﻪ‬
‫ﺍﻗﻠﻴﻢ ﺩﻳﮕﺮ ﻭ ﻳﺎ ﺣﺴﺎﺱﺗﺮ ﺑﻮﺩﻥ ﻓﺮﺍﻳﻨﺪ ﺗﺒﺨﻴﺮ ﺑﻪ ﭘﺎﺭﻩﺍﻱ ﺍﺯ ﻋﻮﺍﻣﻞ ﻳـﺎﺩ‬
‫ﺷﺪﻩ ﺩﺭ ﻳﻚ ﻧﺎﺣﻴﻪ‪ ،‬ﺍﻧﺠﺎﻡ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﺭﻭﻱ ﻣﺪﻝ ‪ ANN‬ﻣﻲﺗﻮﺍﻧـﺪ‬
‫ﻣﻔﻴﺪ ﺑﺎﺷﺪ‪.‬‬
‫‪1‬‬
‫‪2‬‬
‫‪6‬‬
‫‪Adaptive Neuro-Fuzzy‬‬ ‫)‪Inference System (ANFIS‬‬
‫‪7‬‬
‫‪Deswal‬‬
‫‪8‬‬
‫‪Pal‬‬
‫‪9‬‬
‫)‪Hargreaves (HG‬‬
‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬
‫‪ -۱‬ﻣﻘﺪﻣﻪ‬
‫ﻋﻤﺪﻩﺗﺮﻳﻦ ﻣﻌﻴﺎﺭ ﺑﺮﺍﻱ ﺗﻌﻴﻴﻦ ﺩﺭﺟﻪ ﺧﺸﮑﻲ ﺩﺭ ﻳﮏ ﻣﻨﻄﻘﻪ‪ ،‬ﺭﺍﺑﻄﻪ ﺑـﻴﻦ‬
‫ﻣﻘﺪﺍﺭ ﺑﺎﺭﻧﺪﮔﻲ ﺳﺎﻻﻧﻪ ﻭ ﺗﻮﺍﻥ ﺗﺒﺨﻴﺮ ﻣﺤـﻴﻂ ﺍﺳـﺖ‪ .‬ﻫـﺮ ﺍﻧـﺪﺍﺯﻩ ﻣﻘـﺪﺍﺭ‬
‫ﺑﺎﺭﺍﻥ ﻧﺴﺒﺖ ﺑﻪ ﺗﺒﺨﻴﺮ ﮐﻤﺘـﺮ ﺑﺎﺷـﺪ‪ ،‬ﺩﺭﺟـﻪ ﺧﺸـﮑﻲ ﺁﻥ ﻣﻨﻄﻘـﻪ ﺑﻴﺸـﺘﺮ‬
‫ﺍﺳﺖ‪ .‬ﺑﻪﺟﺰ ﻧﻮﺍﺣﻲ ﻣﺤﺪﻭﺩﻱ ﺩﺭ ﺍﻳﺮﺍﻥ )ﺣﺎﺷﻴﻪ ﺩﺭﻳﺎﻱ ﺧﺰﺭ( ﺩﺭ ﺳـﺎﻳﺮ‬
‫ﻣﻨﺎﻃﻖ‪ ،‬ﺗﻮﺍﻥ ﺗﺒﺨﻴﺮ ﺑﻪﻣﺮﺍﺗﺐ ﺑﺎﻻﺗﺮ ﺍﺯ ﻣﻴﺰﺍﻥ ﺑﺎﺭﻧﺪﮔﻲ ﺍﺳﺖ‪ .‬ﺩﺭ ﺍﻳـﺮﺍﻥ‬
‫ﺣﺪﻭﺩ ‪ ۷۲‬ﺩﺭﺻﺪ ﺍﺯ ﻧﺰﻭﻻﺕ ﺟﻮﻱ ﺭﻭﻱ ﺧﺸﮑﻲﻫـﺎ ﺑﻼﻓﺎﺻـﻠﻪ ﺗﺒﺨﻴـﺮ‬
‫ﻣﻲﺷﻮﺩ‪ .‬ﺩﺭ ﻣﻘﻴـﺎﺱ ﺟﻬـﺎﻧﻲ ﺍﻳـﻦ ﺭﻗـﻢ ‪ ۵۷‬ﺩﺭﺻـﺪ ﺑـﻮﺩﻩ ﻭ ﺍﻳـﻦ ﺧـﻮﺩ‬
‫ﺍﻫﻤﻴــﺖ ﺗﺒﺨﻴــﺮ ﻭ ﺗﻌــﺮﻕ‪ ،‬ﺧﺼﻮﺻــﺎﹰ ﺩﺭ ﻣﻨــﺎﻃﻖ ﺑــﺎ ﺍﻗﻠــﻴﻢ ﺧﺸــﮏ ﻭ‬
‫ﻧﻴﻤﻪﺧﺸﮏ ﺍﻳﺮﺍﻥ ﺭﺍ ﻣﺸﺨﺺ ﻣﻲﮐﻨﺪ]‪ .[۱‬ﺩﺭ ﻃﺒﻘﻪﺑﻨﺪﻱ ﺍﻗﻠﻴﻤﻲ ﺍﻳﺮﺍﻥ‪،‬‬
‫ﺷﻬﺮﻫﺎﻱ ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ ﺟﺰﺀ ﻣﻨﺎﻃﻖ ﻧﻴﻤﻪﺧﺸﮏ ﻣﺤﺴﻮﺏ ﻣـﻲﺷـﻮﻧﺪ‪.‬‬
‫ﺑﻪ ﻫﻤﻴﻦ ﺩﻟﻴﻞ ﭘﻴﺶﺑﻴﻨﻲ ﻭ ﺗﻌﻴـﻴﻦ ﻣﻬـﻢﺗـﺮﻳﻦ ﻋﻮﺍﻣـﻞ ﻣـﺆﺛﺮ ﺩﺭ ﻣﻴـﺰﺍﻥ‬
‫ﺷﺪﺕ ﺗﺒﺨﻴﺮ ﺩﺭ ﺍﻳﻦ ﻣﻨﺎﻃﻖ ﭘﺮ ﺍﻫﻤﻴﺖ ﺍﺳـﺖ‪ .‬ﭘﻴﭽﻴـﺪﮔﻲ ﺳﻴﺴـﺘﻢﻫـﺎﻱ‬
‫ﻫﻴــﺪﺭﻭﻟﻮﮊﻳﮑﻲ ﺍﺯ ﻧﻈــﺮ ﺗﻌﻴــﻴﻦ ﺗﻤــﺎﻣﻲ ﭘﺎﺭﺍﻣﺘﺮﻫــﺎﻱ ﺩﺧﻴــﻞ ﻭ ﻧﻘــﺺ‬
‫ﺍﻃﻼﻋﺎﺕ ﺁﻣـﺎﺭﻱ‪ ،‬ﺍﻣﮑـﺎﻥ ﻣـﺪﻝﺳـﺎﺯﻱ ﻓﻴﺰﻳﮑـﻲ ﮐﺎﻣـﻞ ﺳﻴﺴـﺘﻢﻫـﺎﻱ‬
‫ﻫﻴﺪﺭﻭﻟﻮﮊﻳﮑﻲ ﺭﺍ ﻏﻴﺮﻣﻤﮑﻦ ﻣﻲﺳـﺎﺯﺩ‪ .‬ﺩﺭ ﭼﻨـﻴﻦ ﺷـﺮﺍﻳﻄﻲ ﺍﺳـﺘﻔﺎﺩﻩ ﺍﺯ‬
‫ﻣﺪﻝﺳﺎﺯﻱ ﺳﻴﺴﺘﻤﻲ ﮐﻪ ﻣﺒﺘﻨﻲ ﺑﺮ ﺭﻭﺍﺑـﻂ ﺭﻳﺎﺿـﻲ ﺑﺎﺷـﺪ‪ ،‬ﻣـﻮﺭﺩ ﺗﻮﺟـﻪ‬
‫ﻗﺮﺍﺭ ﻣﻲﮔﻴﺮﺩ‪.‬‬
‫ﺍﺯ ﺟﻤﻠﻪ ﺗﮑﻨﻴﮏﻫﺎﻱ ﻣﺪﻝﺳـﺎﺯﻱ ﺑـﻪﺻـﻮﺭﺕ ﺳﻴﺴـﺘﻤﻲ ﻳـﺎ ﺟﻌﺒـﻪ‬
‫‪٢‬‬ ‫‪١‬‬
‫ﺍﺑﺰﺍﺭﻫﺎﻳﻲ ﻗـﻮﻱ ﺩﺭ ﻳـﺎﺩﮔﻴﺮﻱ ﺭﺍﺑﻄـﻪ ﻧﺎﺷـﻨﺎﺧﺘﻪ ﻭ ﭘﻴﭽﻴـﺪﻩ ﺑـﻴﻦ ﻳـﻚ‬
‫ﻓﻀﺎﻱ ﻭﺭﻭﺩﻱ ﺑﻪ ﻓﻀﺎﻱ ﺧﺮﻭﺟﻲ ﻫﺴﺘﻨﺪ ﮐـﻪ ﺩﺭ ﺯﻣﻴﻨـﻪﻫـﺎﻱ ﻣﺨﺘﻠـﻒ‬
‫ﻋﻠﻤﻲ ﺑﻪﮐﺎﺭ ﺑﺮﺩﻩ ﻣﻲﺷﻮﻧﺪ]‪ ۲‬ﻭ ‪ .[۳‬ﺍﻣـﺎ ﺑـﺮﺧﻼﻑ ﻣـﺪﻝﻫـﺎﻱ ﺭﻳﺎﺿـﻲ‬
‫ﺑﻪﺧﻮﺩﻱ ﺧﻮﺩ ﺍﺛﺮ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺑﻪ ﺧﺮﻭﺟﻲ ﻭ ﭼﮕﻮﻧﮕﻲ ﺑﻪﺩﺳﺖ‬
‫ﺁﻣﺪﻥ ﺧﺮﻭﺟﻲ ﺭﺍ ﺗﻮﺿﻴﺢ ﻧﻤـﻲﺩﻫﻨـﺪ‪ ،‬ﺑـﻪﻫﻤـﻴﻦ ﺩﻟﻴـﻞ ﺍﺯ ﺁﻧﺠـﺎﻳﻲ ﮐـﻪ‬
‫ﺍﺭﺯﺵ ﺭﻭﺷﻬﺎﻱ ﻫﻮﺵ ﻣﺼﻨﻮﻋﻲ ﺑﺎ ﺗﻔﺴـﻴﺮ ﻧﺘـﺎﻳﺞ ﺧﺮﻭﺟـﻲ ﺁﻧﻬـﺎ )ﺍﺛـﺮ‬
‫ﻭﺭﻭﺩﻱﻫﺎ ﺑﺮ ﺧﺮﻭﺟﻲ( ﺑﻴﺸﺘﺮ ﻣـﻲﺷـﻮﺩ‪ ،‬ﻣﻄﺎﻟﻌـﺎﺕ ﻣﺘﻌـﺪﺩﻱ ﺑـﺮ ﺭﻭﻱ‬
‫ﺗﻮﺿﻴﺢ ﺭﺍﺑﻄﻪ ﺣﺎﻛﻢ ﺑﺮ ﻣﺤﻴﻂ ﻳﻚ ﺷﺒﻜﻪ ﻋﺼﺒﻲ ﻭ ﻫﻤﭽﻨﻴﻦ ﭼﮕـﻮﻧﮕﻲ‬
‫ﺍﺛﺮ ﭘﺎﺭﺍﻣﺘﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺑﺮ ﺧﺮﻭﺟﻲ ﺍﻧﺠﺎﻡ ﺷﺪﻩ ﺍﺳـﺖ ﮐـﻪ ﺍﺯ ﺁﻥ ﺟﻤﻠـﻪ‬
‫ﻣﻲ ﺗﻮﺍﻥ ﺑﻪ ﻣﻄﺎﻟﻌـﻪ ﻟـﻮ ﻭ ﻫﻤﻜـﺎﺭﺍﻥ‪ ٣‬ﺩﺭ ﺳـﺎﻝ ‪ ۲۰۰۱‬ﺍﺷـﺎﺭﻩ ﻛـﺮﺩ]‪.[۴‬‬
‫ﻧﻤﻮﻧـﻪﺍﻱ ﺍﺯ ﺍﻳـﻦ ﺭﻭﺷـﻬﺎ ﻋﺒـﺎﺭﺕﺍﻧـﺪ ﺍﺯ ﺍﺳـﺘﺨﺮﺍﺝ ﺭﻭﺍﺑـﻂ ﺣـﺎﻛﻢ ﺑـﺮ‬
‫‪٤‬‬
‫ﺭﻭﻱ ﻣﺪﻝ ﺷﺒﮑﻪ ﭘﺲﺍﻧﺘﺸﺎﺭ ﺧﻄﺎ‪ ،٥‬ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﻭﺯﻥﻫـﺎﻱ ﻻﻳـﻪ ﻣﻴـﺎﻧﻲ‬
‫ﻣﺘﺼﻞ ﺑﻪ ﻫﺮ ﻭﺭﻭﺩﻱ ﻭ ﻫﻤﭽﻨﻴﻦ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﺗﺮﻛﻴـﺐ ﻣﻨﻄـﻖ ﻓـﺎﺯﻱ ﺑـﺎ‬
‫ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼﺒﻲ]‪ .[۸-۵‬ﻟﻮ ﻭ ﻫﻤﻜﺎﺭﺍﻥ ﺑﺎ ﺑﺮﺭﺳﻲ ﺍﻳﻦ ﺭﻭﺷﻬﺎ ﺑـﻪ ﺍﻳـﻦ‬
‫ﻧﺘﻴﺠﻪ ﺭﺳﻴﺪﻩﺍﻧﺪ ﻛﻪ ﺍﺛﺮ ﻫﺮ ﭘﺎﺭﺍﻣﺘﺮ ﻭﺭﻭﺩﻱ ﺑﺮ ﺭﻭﻱ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺧﺮﻭﺟﻲ‬
‫‪Black-Box‬‬
‫)‪Artificial Neural Networks(ANNs‬‬
‫‪3‬‬
‫‪Lu et al.‬‬
‫‪4‬‬
‫)‪Principal Component Analysis (PCA‬‬
‫‪5‬‬
‫)‪Back Propagation Neural Networks (BPNN‬‬
‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬
 ‫ﺷــﻬﺮ ﺗﺒﺮﻳــﺰ ﺑﻴﺸــﺘﺮ ﺍﺯ ﺍﺭﻭﻣﻴــﻪ ﻭ ﻣﻴــﺎﻧﮕﻴﻦ ﺭﻃﻮﺑــﺖ ﻧﺴــﺒﻲ‪ ،‬ﺗﺸﻌﺸــﻊ‬
‫ﺧﻮﺭﺷﻴﺪﻱ ﻭ ﻓﺸﺎﺭ )ﺗﺎ ﺣﺪﻭﺩﻱ( ﺩﺭ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ ﺑﻴﺸﺘﺮ ﺍﺯ ﺷـﻬﺮ ﺗﺒﺮﻳـﺰ‬
‫ﺍﺳﺖ‪ .‬ﻣﻘﺎﺩﻳﺮ ﺍﻧﺤﺮﺍﻑ ﺍﺯ ﻣﻌﻴﺎﺭ ﮐﻤ‪‬ﻴﺘﻬﺎﻱ ﺗﺒﺨﻴﺮ‪ ،‬ﺩﻣﺎ‪ ،‬ﺭﻃﻮﺑﺖ ﻭ ﺑـﺎﺩ ﺩﺭ‬
‫ﺷﻬﺮ ﺗﺒﺮﻳﺰ ﺑﻴﺸﺘﺮ ﺍﺯ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ ﻭ ﺍﻧﺤﺮﺍﻑ ﺍﺯ ﻣﻌﻴﺎﺭ ﮐﻤ‪‬ﻴﺘﻬﺎﻱ ﺗﺸﻌﺸـﻊ‬
‫ﺧﻮﺭﺷﻴﺪﻱ ﻭ ﻓﺸﺎﺭ ﺩﺭ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ ﺑﻴﺸﺘﺮ ﺍﺯ ﺗﺒﺮﻳﺰ ﺍﺳﺖ‪.‬‬
‫ﻭ ﺗﺤﻘﻴﻘﺎﺕ ﺻﻮﺭﺕ ﮔﺮﻓﺘﻪ ﺩﺭ ﺯﻣﻴﻨﻪ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﻭ ‪ ANNs‬ﻧﺘﻴﺠـﻪ‬
‫‪ -۳‬ﺭﻭﺷﻬﺎﻱ ﺑﺮﺁﻭﺭﺩ ﺗﺒﺨﻴﺮ‬
‫ﭼﺎﻭ ﻭ ﻫﻤﮑﺎﺭﺍﻥ‪ ٨‬ﺳﻪ ﺭﻭﺵ ﺗﺠﺮﺑﻲ ﺑﺮﺍﻱ ﺗﺨﻤﻴﻦ ﺗﺒﺨﻴـﺮ ﺍﺯ ﺳـﻄﺢ ﺁﺯﺍﺩ‬
‫ﺁﺑﻬﺎ ﺭﺍ ﺍﺭﺍﺋـﻪ ﮐـﺮﺩﻩﺍﻧـﺪ ﮐـﻪ ﻋﺒـﺎﺭﺕﺍﻧـﺪ ﺍﺯ‪ :‬ﺭﻭﺷـﻬﺎﻱ ﺁﻳﺮﻭﺩﻳﻨـﺎﻣﻴﻜﻲ‪،٩‬‬
‫ﺗﻌﺎﺩﻝ ﺍﻧﺮﮊﻱ‪ ١٠‬ﻭ ﺗﺮﻛﻴﺒﻲ ﭘـﻨﻤﻦ‪ .١١‬ﺍﻳـﻦ ﺭﻭﺷـﻬﺎﻱ ﺗﺠﺮﺑـﻲ ﺑـﻪﻣﻨﻈـﻮﺭ‬
‫ﺑﺮﺁﻭﺭﺩ ﻣﻘﺎﺩﻳﺮ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﺩﺭ ﻣﻄﺎﻟﻌﻪ ﺣﺎﺿﺮ ﺑـﻪﮐـﺎﺭ ﮔﺮﻓﺘـﻪ ﺷـﺪ ﮐـﻪ‬
‫ﺗﻮﺿﻴﺢ ﺟﺎﻣﻊ ﺩﺭ ﻣﻮﺭﺩ ﻫﺮ ﺳـﻪ ﺭﻭﺵ ﺩﺭ ﻣﻄﺎﻟﻌـﺎﺕ ﻗﺒﻠـﻲ ﺁﻭﺭﺩﻩ ﺷـﺪﻩ‬
‫ﺍﺳﺖ]‪ .[۲۰‬ﺍﺯ ﺟﻤﻠـﻪ ﻣـﺪﻝﻫـﺎﻱ ﺟﻌﺒـﻪ ﺳـﻴﺎﻩ‪ ،‬ﻣـﺪﻝ ﺭﮔﺮﺳـﻴﻮﻥ ﺧﻄـﻲ‬
‫ﭼﻨﺪﻣﺘﻐﻴﺮﻩ‪ ١٢‬ﻭ ﻣﺪﻝﻫﺎﻱ ‪ ANN‬ﺭﺍ ﻣﻲ ﺗﻮﺍﻥ ﻧﺎﻡ ﺑﺮﺩ ﮐﻪ ﺩﺭ ﺍﻳﻦ ﻣﻄﺎﻟﻌـﻪ‬
‫ﻣﻮﺭﺩ ﺍﺳـﺘﻔﺎﺩﻩ ﻗـﺮﺍﺭ ﮔﺮﻓـﺖ]‪۲۱‬ﻭ‪ .[۲۲‬ﺭﺍﺑﻄـﻪ ﺭﮔﺮﺳـﻴﻮﻥ ﭼﻨـﺪﻣﺘﻐﻴﺮﻩ‬
‫ﺑﻪﺻﻮﺭﺕ ﮐﻠﻲ ﺯﻳﺮ ﺍﺳﺖ]‪[۲۳‬‬
‫‪Y  0  1X1   2 X 2        K X K  A‬‬ ‫)‪(۱‬‬
‫ﮐﻪ ﺩﺭ ﺍﻳﻦ ﺭﺍﺑﻄﻪ‬
‫‪ Y‬ﻣﺘﻐﻴﺮ ﻭﺍﺑﺴﺘﻪ‪ X1 ،‬ﺗـﺎ ‪ XK‬ﻣﺘﻐﻴﺮﻫـﺎﻱ ﻣﺴـﺘﻘﻞ‪ ،‬ﭘﺎﺭﺍﻣﺘﺮﻫـﺎﻱ ‪ 0‬ﺗـﺎ‬
‫‪ K‬ﺿﺮﺍﻳﺐ ﻭ‪ A‬ﺧﻄﺎﻱ ﺭﮔﺮﺳﻴﻮﻥ ﺍﺳﺖ‪ .‬ﺍﻳﻦ ﻣﺪﻝ ﻣﻮﻗﻌﻴﺖ ﻳـﮏ ﺍﺑـﺮ‬
‫ﺻﻔﺤﻪ ‪ ١٣‬ﺩﺭ ﻓﻀﺎﻱ ‪ K‬ﺑﻌﺪﻱ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻣﺴﺘﻘﻞ ﺭﺍ ﺗﻮﺻـﻴﻒ ﻣـﻲﮐﻨـﺪ‪.‬‬
‫ﭘﺮﺍﮐﻨﺶ ﻣﺸﺎﻫﺪﺍﺕ ﺩﺭ ﻳﮏ ﻓﻀﺎﻱ ‪ K‬ﺑﻌﺪﻱ ﺑـﻮﺩﻩ ﻭ ﻣﻮﻗﻌﻴـﺖ ﺑﻬﺘـﺮﻳﻦ‬
‫ﺍﺑﺮ ﺻﻔﺤﻪ ﺭﮔﺮﺳﻴﻮﻥ ﮐﻪ ﻣﻲﺗﻮﺍﻥ ﺑﻴﻦ ﺁﻧﻬﺎ ﺑﺮﺍﺯﺵ ﺩﺍﺩ‪ ،‬ﺗﻌﻴﻴﻦ ﻣـﻲﺷـﻮﺩ‪.‬‬
‫ﺩﺭ ﺭﺍﺑﻄﻪ ﺭﮔﺮﺳﻴﻮﻥ ﺑـﺮﺍﻱ ﻳـﺎﻓﺘﻦ ﺑﻬﺘـﺮﻳﻦ ﺑـﺮﺍﺯﺵ ﻣـﺪﻝ ﺑـﺮ ﺩﺍﺩﻩﻫـﺎﻱ‬
‫‪١٤‬‬
‫ﻣﺸــﺎﻫﺪﺍﺗﻲ‪ ،‬ﭘﺎﺭﺍﻣﺘﺮﻫــﺎﻱ ﺭﺍﺑﻄــﻪ ﺑــﻪﺭﻭﺵ ﮐﻤﺘــﺮﻳﻦ ﻣﺮﺑﻌــﺎﺕ ﺧﻄــﺎ‬
‫ﻣﺤﺎﺳﺒﻪ ﻣﻲﺷﻮﻧﺪ‪.‬‬
‫‪8‬‬
‫‪Chow et al.‬‬
‫‪9‬‬
‫‪Aerodynamic‬‬
‫‪10‬‬
‫‪Energy Balance‬‬
‫‪11‬‬
‫‪Penman‬‬
‫‪12‬‬
‫)‪Multiple Linear Regression(MLR‬‬
‫‪6‬‬
‫‪13‬‬
‫‪Hyperplane‬‬
‫‪14‬‬
‫‪Least Squares Error‬‬
‫ﺩﺭ ﻋﻠﻮﻡ ﻣﺨﺘﻠﻒ‪ ،‬ﻣﻄﺎﻟﻌـﺎﺗﻲ ﺩﺭ ﺯﻣﻴﻨـﻪ ﺁﻧـﺎﻟﻴﺰ ﺣﺴﺎﺳـﻴﺖ ‪ANNs‬‬
‫ﺑﺮﺍﻱ ﻣﺸﺨﺺ ﮐﺮﺩﻥ ﻣﻴﺰﺍﻥ ﺍﺛﺮﭘﺬﻳﺮﻱ ﺧﺮﻭﺟﻲ ﺷـﺒﮑﻪ ﺍﺯ ﻭﺭﻭﺩﻱﻫـﺎﻱ‬
‫ﻣﺨﺘﻠﻒ ﺁﻥ ﺑﻪﺭﻭﺷﻬﺎﻱ ﮔﻮﻧـﺎﮔﻮﻥ ﺍﻧﺠـﺎﻡ ﺷـﺪﻩ ﺍﺳـﺖ‪ ،‬ﻣﺎﻧﻨـﺪ ﺗﺤﻘﻴﻘـﺎﺕ‬
‫ﺍﻧﺠﺎﻡ ﺷﺪﻩ ﺗﻮﺳﻂ ﺑﻨﻲﻣﻬﺪ‪ ١‬ﻭ ﻳﺜﺮﻭﺑـﻲ‪ ،٢‬ﺟـﻴﻦ ﻭ ﻫﻤﮑـﺎﺭﺍﻥ‪ ،٣‬ﺟـﻮﺭﻱ ﻭ‬
‫ﻫﻤﮑﺎﺭﺍﻥ‪ ٤‬ﻭ ﻟﻴﻮﻧﻖ ﻭ ﻫﻤﮑﺎﺭﺍﻥ‪ .[۱۹-۱۶]٥‬ﺍﺯ ﺑﺮﺭﺳﻲ ﺍﺩﺑﻴﺎﺕ ﮔﺬﺷﺘﻪ ﻭ‬
‫ﮔﺮﻓﺘﻪ ﻣﻲﺷﻮﺩ ﮐﻪ ﺗﺤﻘﻴﻘـﻲ ﺩﺭ ﺧﺼـﻮﺹ ﺗﺤﻠﻴـﻞ ﺣﺴﺎﺳـﻴﺖ ﺧﺼﻮﺻـﺎﹰ‬
‫ﺑﻪﺭﻭﺵ ﻣﺸﺘﻘﺎﺕ ﺟﺰﺋـﻲ ﺩﺭ ﺯﻣﻴﻨـﻪ ﮐﺎﺭﻫـﺎﻱ ﻫﻴـﺪﺭﻭﻟﻮﮊﻳﮑﻲ ﻭ ﻓﺮﺍﻳﻨـﺪ‬
‫ﺗﺒﺨﻴــﺮ ﺻــﻮﺭﺕ ﻧﮕﺮﻓﺘــﻪ ﺍﺳــﺖ‪ .‬ﺩﺭ ﻣﻘﺎﻟــﻪ ﺣﺎﺿــﺮ ﺑــﻪ ﺑﺮﺭﺳــﻲ ﻗــﺪﺭﺕ‬
‫‪ ANNs‬ﺩﺭ ﺑﺮﺁﻭﺭﺩ ﻣﻴﺰﺍﻥ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﺗﺸﺘﮏ ﺩﺭ ﻣﻨﺎﻃﻖ ﻣﻄﺎﻟﻌﺎﺗﻲ ﻭ‬
‫ﺗﺤﻠﻴﻞ ﺣﺴﺎﺳﻴﺖ ﻓﺮﺍﻳﻨـﺪ ﻣـﻮﺭﺩ ﻧﻈـﺮ ﻧﺴـﺒﺖ ﺑـﻪ ﻋﻮﺍﻣـﻞ ﻣـﺆﺛﺮ ﺩﺭ ﺁﻥ‬
‫ﭘﺮﺩﺍﺧﺘﻪ ﺷﺪ‪.‬‬
‫‪ -۲‬ﻣﻨﺎﻃﻖ ﻣﻄﺎﻟﻌﺎﺗﻲ ﻭ ﺩﺍﺩﻩﻫﺎ‬
‫ﺩﺭ ﺍﻳﻦ ﺗﺤﻘﻴﻖ ﺗﺒﺨﻴﺮ ﺷﻬﺮﻫﺎﻱ ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ ﻭﺍﻗـﻊ ﺩﺭ ﺷـﻤﺎﻝ ﻏﺮﺑـﻲ‬
‫ﺍﻳﺮﺍﻥ ﻣـﺪﻝ ﺷـﺪﻩ ﮐـﻪ ﺑـﺮﺍﻱ ﺍﻳـﻦ ﻣﻨﻈـﻮﺭ ﺍﺯ ﺩﺍﺩﻩﻫـﺎﻱ ﻫﻴـﺪﺭﻭﻟﻮﮊﻳﮑﻲ‬
‫ﻭ ﺭﻭﺯﺍﻧــﻪ ﺍﻳﺴــﺘﮕﺎﻫﻬﺎﻱ ﺳــﻴﻨﻮﭘﺘﻴﮏ ﺗﺒﺮﻳــﺰ ﺩﺭ ﻣﺤــﺪﻭﺩﻩ ﺳــﺎﻟﻬﺎﻱ‬
‫‪ ۱۳۸۷-۱۳۷۳‬ﻭ ﺍﺭﻭﻣﻴــﻪ ﺩﺭ ﺳــﺎﻟﻬﺎﻱ ‪ ۱۳۸۱-۱۳۷۴‬ﮐــﻪ ﺑــﻪﺗﺮﺗﻴــﺐ‬
‫ﺷﺎﻣﻞ ‪ ۵۴۷۹‬ﻭ ‪ ۲۹۲۲‬ﺭﮐﻮﺭﺩ ﺛﺒـﺖ ﺷـﺪﻩ ﺑـﺮﺍﻱ ﻫـﺮ ﭘـﺎﺭﺍﻣﺘﺮ ﺑﻮﺩﻧـﺪ‪،‬‬
‫ﺍﺳﺘﻔﺎﺩﻩ ﺷﺪ‪ .‬ﻣﺘﻐﻴﺮﻫﺎ ﻋﺒﺎﺭﺕ ﺑﻮﺩﻧﺪ ﺍﺯ‪ :‬ﺗﺸﻌﺸﻊ ﺧﻮﺭﺷﻴﺪ‪ ، ٦‬ﺩﻣﺎﻱ ﻫـﻮﺍ‬
‫)‪ ، (T‬ﺭﻃﻮﺑﺖ ﻧﺴﺒﻲ‪ ، ٧‬ﺳﺮﻋﺖ ﺑﺎﺩ ‪ ، U2‬ﻓﺸﺎﺭ ﺳﺘﻮﻥ ﻫﻮﺍ )‪ (P‬ﻭ ﺗﺒﺨﻴـﺮ‬
‫ﺗﺸﺘﮏ ‪ Et‬ﮐﻪ ﺑﻪﻣﻨﻈﻮﺭ ﻣﺪﻝﺳﺎﺯﻱ ﺑﺎ ‪ ANNs‬ﺑـﻪ ﺩﻭ ﻗﺴـﻤﺖ ﺁﻣﻮﺯﺷـﻲ‬
‫)‪ ۷۵‬ﺩﺭﺻﺪ ﺩﺍﺩﻩﻫـﺎ( ﻭ ﺁﺯﻣﺎﻳﺸـﻲ )‪ ۲۵‬ﺩﺭﺻـﺪ ﺩﺍﺩﻩﻫـﺎ( ﺗﻘﺴـﻴﻢﺑﻨـﺪﻱ‬
‫ﺷﺪﻧﺪ‪ .‬ﺟﺪﻭﻝ ‪ ۱‬ﺑﺮﺧﻲ ﺍﺯ ﻣﺸﺨﺼﺎﺕ ﺁﻣﺎﺭﻱ ﺩﺍﺩﻩﻫﺎ ﺭﺍ ﺑـﺮﺍﻱ ﺩﻭ ﺩﺳـﺘﻪ‬
‫ﺁﻣﻮﺯﺷﻲ ﻭ ﺁﺯﻣﺎﻳﺸﻲ ﮐﻪ ﺩﺭ ﻣﺪﻝ ﺷﺒﮑﻪ ﺍﺳﺘﻔﺎﺩﻩ ﺷﺪ ﺭﺍ ﻧﺸﺎﻥ ﻣـﻲﺩﻫـﺪ‪.‬‬
‫ﺑﺮﺍﻱ ﺩﻭﺭﻩﻫﺎﻱ ﺁﻣﺎﺭﻱ ﺩﺭﻧﻈﺮ ﮔﺮﻓﺘﻪ ﺷﺪﻩ ﻣﻴﺎﻧﮕﻴﻦ ﺗﺒﺨﻴﺮ‪ ،‬ﺩﻣﺎ ﻭ ﺑـﺎﺩ ﺩﺭ‬
‫‪1‬‬
‫‪Banimahd‬‬
‫‪2‬‬
‫‪Yasrobi‬‬
‫‪3‬‬
‫‪Jain et al.‬‬
‫‪4‬‬
‫‪Gevrey et al.‬‬
‫‪5‬‬
‫‪Liong et al.‬‬
‫)‪Solar Radiation (R a‬‬
‫‪7‬‬ ‫‪(R‬‬‫)‬
‫‪Relative Humidity‬‬ ‫‪H‬‬

‫ﺟﺪﻭﻝ ‪ -۱‬ﻣﺸﺨﺼﺎﺕ ﺁﻣﺎﺭﻱ ﮔﺮﻭﻫﻬﺎﻱ ﺁﻣﻮﺯﺷﻲ ﻭ ﺁﺯﻣﺎﻳﺸﻲ‬

‫ﺍﻳﺴﺘﮕﺎﻩ ﺗﺒﺮﻳﺰ‬ ‫ﺍﻳﺴﺘﮕﺎﻩ ﺍﺭﻭﻣﻴﻪ‬
‫ﻣﺘﻐﻴﺮﻫﺎ‬
‫ﮔﺮﻭﻩ ﺁﻣﻮﺯﺷﻲ‬ ‫ﮔﺮﻭﻩ ﺁﺯﻣﺎﻳﺸﻲ‬ ‫ﮔﺮﻭﻩ ﺁﻣﻮﺯﺷﻲ‬ ‫ﮔﺮﻭﻩ ﺁﺯﻣﺎﻳﺸﻲ‬
‫ﻣﻴﺎﻧﮕﻴﻦ‬ ‫ﻛﻤﻴﻨﻪ ﺑﻴﺸﻴﻨﻪ‬ ‫‪Std‬‬ ‫ﻣﻴﺎﻧﮕﻴﻦ‬ ‫ﻛﻤﻴﻨﻪ ﺑﻴﺸﻴﻨﻪ‬ ‫‪Std‬‬ ‫ﻣﻴﺎﻧﮕﻴﻦ‬ ‫ﻛﻤﻴﻨﻪ ﺑﻴﺸﻴﻨﻪ‬ ‫‪Std‬‬ ‫ﻣﻴﺎﻧﮕﻴﻦ‬ ‫ﻛﻤﻴﻨﻪ ﺑﻴﺸﻴﻨﻪ‬ ‫‪Std‬‬
‫‪5/93‬‬ ‫‪22/1‬‬ ‫‪0/10‬‬ ‫‪5/12‬‬ ‫‪5/85‬‬ ‫‪21/2‬‬ ‫‪0/10‬‬ ‫‪5/19‬‬ ‫‪4/193‬‬ ‫‪13/6‬‬ ‫‪0/011‬‬ ‫‪3/234‬‬ ‫‪4/335‬‬ ‫‪13/2‬‬ ‫‪0/016‬‬ ‫‪3/561‬‬ ‫‪Et‬‬
‫)‪(mm/d‬‬
‫‪13/21‬‬ ‫‪33/17‬‬ ‫‪14/97‬‬ ‫‪10/06‬‬ ‫‪13/12‬‬ ‫‪34/02‬‬ ‫‪12/17‬‬ ‫‪11/07‬‬ ‫‪11/95‬‬ ‫‪28/75‬‬ ‫‪12/95‬‬ ‫‪9/529‬‬ ‫‪10/88‬‬ ‫‪26/97‬‬ ‫‪8/875‬‬ ‫‪9/124‬‬ ‫)℃( ‪T‬‬
‫‪RH‬‬
‫‪50/25‬‬ ‫‪96/12‬‬ ‫‪9/90‬‬ ‫‪17/54‬‬ ‫‪50/34‬‬ ‫‪96/25‬‬ ‫‪14/37‬‬ ‫‪17/14‬‬ ‫‪57/52‬‬ ‫‪100‬‬ ‫‪15/62‬‬ ‫‪15/45‬‬ ‫‪59/31‬‬ ‫‪98/12‬‬ ‫‪28/75‬‬ ‫‪15/62‬‬ ‫)ﺩﺭﺻﺪ(‬
‫‪3/28‬‬ ‫‪9/37‬‬ ‫‪0/25‬‬ ‫‪1/54‬‬ ‫‪3/55‬‬ ‫‪9/62‬‬ ‫‪0/50‬‬ ‫‪1/49‬‬ ‫‪2/176‬‬ ‫‪8/25‬‬ ‫‪0/1‬‬ ‫‪1/100‬‬ ‫‪1/614‬‬ ‫‪7/375‬‬ ‫‪0/1‬‬ ‫‪1/212‬‬ ‫)‪U2 (m/s‬‬
‫‪386/3‬‬ ‫‪894/0‬‬ ‫‪20/40‬‬ ‫‪194/7‬‬ ‫‪354/6‬‬ ‫‪791/0‬‬ ‫‪24/84‬‬ ‫‪178/4‬‬ ‫‪443/9‬‬ ‫‪950/2‬‬ ‫‪11/07‬‬ ‫‪184/6‬‬ ‫‪401/0‬‬ ‫‪943/1‬‬ ‫‪38‬‬ ‫‪231/9‬‬ ‫‪Ra‬‬
‫)‪(w/m2‬‬
‫‪864/2‬‬ ‫‪878/2‬‬ ‫‪848/3‬‬ ‫‪4/32‬‬ ‫‪864/4‬‬ ‫‪879/7‬‬ ‫‪851/8‬‬ ‫‪4/41‬‬ ‫‪867/3‬‬ ‫‪881/1‬‬ ‫‪851/3‬‬ ‫‪4/483‬‬ ‫‪868/0‬‬ ‫‪881/4‬‬ ‫‪854/9‬‬ ‫‪4/451‬‬ ‫)‪P (Hpa‬‬

‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫‪ ‬‬

‫ﻣﻴﺎﻧﻲ ﺍﺯ ‪ ۱‬ﺗﺎ ‪ ۱۰‬ﻭ ﺑﺮﺍﻱ ﺩﻭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ﺑﻪﺻﻮﺭﺕ ﺗﺼﺎﺩﻓﻲ ﺑﺎﺯﻩ ﺍﻋـﺪﺍﺩ‬ ‫‪ -۱-۳‬ﻣﺪﻝ ﺷﺒﮑﻪ ﻋﺼﺒﻲ ﻣﺼﻨﻮﻋﻲ‬
‫ﺑﻴﻦ ‪ ۵‬ﺗﺎ ‪ ۱۲‬ﺍﻣﺘﺤﺎﻥ ﺷﺪ ﺗﺎ ﺑﻬﺘﺮﻳﻦ ﺣﺎﻟـﺖ ﺑـﺮ ﺍﺳـﺎﺱ ﺁﺯﻣـﻮﻥ ﻭ ﺧﻄـﺎ‬ ‫ﺍﺯ ﺟﻤﻠﻪ ﺷﺒﻜﻪﻫﺎﻱ ﻋﺼـﺒﻲ ﮐـﻪ ﺑـﻪﻃـﻮﺭ ﻭﺳـﻴﻌﻲ ﺩﺭ ﻋﻠـﻮﻡ ﻣﺨﺘﻠـﻒ ﺍﺯ‬
‫ﺑﻪ ﺩﺳﺖ ﺁﻳﺪ‪ .‬ﺑـﺮﺍﻱ ﺍﺭﺯﻳـﺎﺑﻲ ﺭﻭﺷـﻬﺎﻱ ﻳـﺎﺩ ﺷـﺪﻩ ﺷﺎﺧﺼـﻬﺎﻱ ﺁﻣـﺎﺭﻱ‬ ‫ﺟﻤﻠﻪ ﻣﻬﻨﺪﺳﻲ ﺁﺏ ﻣﻮﺭﺩ ﺍﺳﺘﻔﺎﺩﻩ ﻗﺮﺍﺭ ﮔﺮﻓﺘﻪ ﺍﺳﺖ‪ BPNN ،‬ﺍﺳـﺖ ﻛـﻪ‬
‫ﺿـﺮﻳﺐ ﺗﺒﻴــﻴﻦ‪ (R2) ٥‬ﻭ ﺟــﺬﺭ ﻣﺘﻮﺳــﻂ ﻣﺮﺑﻌـﺎﺕ ﺧﻄــﺎ‪ ٦‬ﺑــﻪﮐــﺎﺭ ﺑــﺮﺩﻩ‬ ‫ﺩﺭ ﺍﻳﻦ ﻣﻄﺎﻟﻌﻪ ﻫﻤﺮﺍﻩ ﺑﺎ ﺍﻟﮕﻮﺭﻳﺘﻢ ﺁﻣﻮﺯﺷﻲ ﻟﻮﻧﺒﺮﮒ ﻣﺎﺭﮔـﺎﺭﺕ‪ ١‬ﺑـﻪﻛـﺎﺭ‬
‫ﺷﺪ]‪.[۲۴‬‬ ‫ﮔﺮﻓﺘﻪ ﺷﺪ‪ .‬ﺍﺯ ﻧﻈﺮ ﺳﺎﺧﺘﺎﺭ ﺭﻳﺎﺿﻲ‪ BPNN ،‬ﻳﮏ ﺍﻟﮕﻮﺭﻳﺘﻢ ﺑﻬﻴﻨﻪﺳـﺎﺯﻱ‬
‫ﺷﻴﺐ ﻧﺰﻭﻟﻲ ﺍﺳﺖ ﮐﻪ ﺑﺎ ﻫﺪﻑ ﻛﻤﻴﻨـﻪ ﮐـﺮﺩﻥ ﺧﻄـﺎﻱ ﮐـﻞ ﺑـﻴﻦ ﻣﻘـﺎﺩﻳﺮ‬
‫‪ -۴‬ﺁﻧﺎﻟﻴﺰﺣﺴﺎﺳﻴﺖ ﻣﺪﻝ‬ ‫ﺧﺮﻭﺟﻲ ﺍﺯ ‪ ANN‬ﻭ ﻣﻘﺎﺩﻳﺮ ﻣﺸﺎﻫﺪﺍﺗﻲ ﺳﺎﺧﺘﻪ ﺷﺪﻩ ﺍﺳﺖ]‪ .[۴‬ﺑﺎ ﺗﻮﺟـﻪ‬
‫ﺭﻭﺷﻬﺎﻱ ﻣﺨﺘﻠﻔﻲ ﺗﺤﺖ ﻧﺎﻡ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳـﻴﺖ ﺑـﺮﺍﻱ ﺍﺳـﺘﺨﺮﺍﺝ ﻗـﻮﺍﻧﻴﻦ‬ ‫ﺑﻪ ﻣﺤﺪﻭﺩﻩ ﺗﻮﺍﺑﻊ ﻣﺤﺮﮎ‪ ،‬ﺍﻧﺘﻘﺎﻝ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺧﺮﻭﺟﻲ ﻭ ﻭﺭﻭﺩﻱ ﺑﻪ ﺑـﺎﺯﻩ‬
‫ﺣﺎﮐﻢ ﺑﺮ ﻣﺪﻝ ﻫﺪﻑ ﺑﻪﮐﺎﺭ ﺑﺮﺩﻩ ﻣﻲﺷـﻮﻧﺪ‪ .‬ﻣﻄﺎﻟﻌـﻪ ﺣﺎﺿـﺮ ﺑـﻪ ﺑﺮﺭﺳـﻲ‬ ‫ﻣﻨﺎﺳﺐ ﺑﺎ ﻳﮏ ﺗﻮﺯﻳﻊ ﻻﺯﻡ ﺑﻪﻧﻈﺮ ﻣـﻲﺭﺳـﺪ‪ .‬ﻣﻬـﻢﺗـﺮﻳﻦ ﻫـﺪﻑ ﺍﺯ ﺍﻳـﻦ‬
‫ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﻣـﺪﻝﻫـﺎﻱ ‪ ANN‬ﺑـﻪ ﺩﻭ ﺭﻭﺵ ﻭﺯﻥﻫـﺎﻱ ﺍﺗﺼـﺎﻝ‪ ٧‬ﻭ‬ ‫ﺍﻧﺘﻘﺎﻝ‪ ،‬ﺗﺼﺤﻴﺢ ﭘﺨﺶ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﻭ ﺧﺮﻭﺟـﻲ ﺑـﻪﻧﺤـﻮﻱ ﺍﺳـﺖ‬
‫ﻣﺸﺘﻘﺎﺕ ﺟﺰﺋﻲ‪ ٨‬ﺧﺮﻭﺟﻲ ﻧﺴﺒﺖ ﺑﻪ ﻭﺭﻭﺩﻱ ﭘﺮﺩﺍﺧﺖ‪ .‬ﺁﻧﺎﻟﻴﺰﺣﺴﺎﺳﻴﺖ‬ ‫ﮐﻪ ﺧﻄﺎﻱ ﻣﺪﻝﺳﺎﺯﻱ ﮐﻢ ﺷﻮﺩ‪ .‬ﺑﺎ ﺗﻮﺟﻪ ﺑﻪ ﺷﮑﻞ ﻭ ﺭﻓﺘﺎﺭ ﺗﻮﺍﺑﻊ ﻣﺤﺮﮎ‪،‬‬
‫ﺑﻪﺭﻭﺵ ﻭﺯﻥﻫﺎﻱ ﺍﺗﺼﺎﻟﻲ ﺑﺮﺍﻱ ﺍﻭﻟﻴﻦ ﺑﺎﺭ ﺗﻮﺳﻂ ﮔﺎﺭﺳﻮﻥ‪ ٩‬ﺑﻪﮐﺎﺭ ﺑـﺮﺩﻩ‬ ‫ﺯﻣﺎﻧﻲ ﮐﻪ ﻭﺭﻭﺩﻱﻫﺎ ﺩﺭ ﻣﺤﺪﻭﺩﻩ )‪ (۰ ، ۱‬ﻳﺎ )‪ (-۱ ، ۱‬ﻗﺮﺍﺭ ﻣـﻲﮔﻴﺮﻧـﺪ‪،‬‬
‫ﺷﺪﻩ ﺍﺳﺖ ﻭ ﺗﻮﺳﻂ ﮔﻮﺡ‪ ١٠‬ﺗﮑﺮﺍﺭ ﺷﺪﻩ ﺍﺳﺖ]‪.[۲۷-۲۵‬‬ ‫ﻣﻴﺰﺍﻥ ﺗﻐﻴﻴﺮﺍﺕ ﺩﺭ ﺧﺮﻭﺟﻲ ﺍﻳﻦ ﺗﻮﺍﺑﻊ ﺑـﻪﺍﺯﺍﻱ ﺗﻐﻴﻴـﺮﺍﺕ ﻭﺭﻭﺩﻱ ﺑﻬﺘـﺮ‬
‫ﻧﻤﺎﻳﺎﻥ ﺷﺪﻩ ﻭ ﺳﺮﻋﺖ ﺁﻣﻮﺯﺵ ﻭ ﻫﻤﮕﺮﺍﻳﻲ ﺑﻴﺸﺘﺮ ﻣﻲﮔﺮﺩﺩ‪ ،‬ﻟﺬﺍ ﺩﺭ ﺍﻳـﻦ‬
‫‪ -۱-۴‬ﺭﻭﺵ ﻣﺸﺘﻘﺎﺕ ﺟﺰﺋﻲ‬ ‫ﻣﻄﺎﻟﻌﻪ ﻗﺒﻞ ﺍﺯ ﻭﺍﺭﺩ ﮐﺮﺩﻥ ﺩﺍﺩﻩﻫﺎ ﺑﻪ ﺷﺒﮑﻪ‪ ،‬ﻣﻘﻴـﺎﺱﺩﻫـﻲ ﺭﻭﻱ ﺩﺍﺩﻩﻫـﺎ‬
‫ﺭﻭﺵ ‪ PaD‬ﺑﻪﺻﻮﺭﺕ ﻣﺸـﺘﻖ ﺟﺰﺋـﻲ ﺧﺮﻭﺟـﻲ ﻧﺴـﺒﺖ ﺑـﻪ ﻭﺭﻭﺩﻱﻫـﺎ‬ ‫ﺍﻋﻤﺎﻝ ﺷﺪ ﺗـﺎ ﺗﻤـﺎﻣﻲ ﺩﺍﺩﻩﻫـﺎ ﺑـﻪﺻـﻮﺭﺕ ﺍﻧﺘﺨـﺎﺑﻲ ﺑـﻴﻦ )‪ (۰ ، ۱‬ﻗـﺮﺍﺭ‬
‫ﺗﻌﺮﻳﻒ ﻣﻲﺷﻮﺩ‪ .‬ﺑﺮﺍﻱ ﻳﮏ ﺷﺒﮑﻪ ﭘﺮﺳﭙﺘﺮﻭﻥ ﺳـﻪ ﻻﻳـﻪ ﻓﺮﺿـﻲ ﮐـﻪ ﺩﺭ‬ ‫ﺑﮕﻴﺮﻧــﺪ‪ .‬ﺑــﻪﻣﻨﻈــﻮﺭ ﻳــﺎﻓﺘﻦ ﺑﻬﺘــﺮﻳﻦ ﻧﮕﺎﺷــﺖ ﻭﺭﻭﺩﻱ‪ -‬ﺧﺮﻭﺟــﻲ‪ ،‬ﺩﺭ‬
‫ﺷﮑﻞ ‪ ۱‬ﻧﺸﺎﻥ ﺩﺍﺩﻩ ﺷﺪﻩ ﺍﺳﺖ‪ ،‬ﻣـﻲﺗـﻮﺍﻥ ﺭﻭﺍﺑـﻂ ﺁﻧـﺎﻟﻴﺰ ﺣﺴﺎﺳـﻴﺖ ﺭﺍ‬ ‫ﺷﺒﮑﻪﻫﺎﻱ ﺁﺯﻣﺎﻳﺸﻲ ﺗﻌـﺪﺍﺩ ﺣـﺪﺍﮐﺜﺮ ﺩﻭ ﻻﻳـﻪ ﻣﻴـﺎﻧﻲ ﺑـﺎ ﺗﻮﺍﺑـﻊ ﺍﻧﺘﻘـﺎﻝ‬
‫ﺑﻪﺩﺳﺖ ﺁﻭﺭﺩ‪.‬‬ ‫ﺗﺎﻧﮋﺍﻧﺖ ﺳﻴﮕﻤﻮﺋﻴﺪ‪ ٢‬ﻭ ﻟﮕﺎﺭﻳﺘﻢ ﺳـﻴﮕﻤﻮﺋﻴﺪ‪ ٣‬ﻭ ﺑـﺮﺍﻱ ﻻﻳـﻪ ﺧﺮﻭﺟـﻲ ﺍﺯ‬
‫ﺗﺎﺑﻊ ﺍﻧﺘﻘﺎﻝ ﺧﻄﻲ‪ ٤‬ﺍﺳﺘﻔﺎﺩﻩ ﺷﺪ‪ .‬ﺑﺮﺍﻱ ﻳـﺎﻓﺘﻦ ﺗﻌـﺪﺍﺩ ﻧـﺮﻭﻥ ﻣﻨﺎﺳـﺐ ﺩﺭ‬
‫ﻻﻳﻪﻫﺎﻱ ﻣﻴﺎﻧﻲ ﺷﺒﮑﻪ‪ ،‬ﺁﺭﺍﻳﺶ ﻧﺮﻭﻥﻫﺎﻱ ﻻﻳﻪ ﻣﻴـﺎﻧﻲ ﺑـﺮﺍﻱ ﺗـﮏ ﻻﻳـﻪ‬
‫‪5‬‬
‫‪Determination Coefficient‬‬
‫‪6‬‬
‫)‪Root Mean Squared Error (RMSE‬‬
‫‪7‬‬
‫‪Weight Method‬‬
‫‪8‬‬ ‫‪1‬‬
‫‪Partial Derivation‬‬ ‫)‪Method (PaD‬‬ ‫‪Levenberg-Marquardt‬‬
‫‪9‬‬ ‫‪2‬‬
‫‪Garson‬‬ ‫‪Tansig‬‬
‫‪10‬‬ ‫‪3‬‬
‫‪Goh‬‬ ‫‪Logsig‬‬
‫‪4‬‬
‫‪Purelin‬‬

‫‪+1‬‬

‫‪S‬‬ ‫∅‬
‫‪+‬‬
‫‪w‬‬

‫‪X‬‬ ‫‪w‬‬ ‫‪+1‬‬ ‫‪N‬‬

‫‪X‬‬ ‫‪S‬‬ ‫∅‬ ‫‪+1‬‬
‫‪+‬‬
‫‪N‬‬
‫‪X‬‬ ‫‪w‬‬ ‫‪S‬‬ ‫‪N‬‬
‫‪+‬‬ ‫∅‬
‫‪+1‬‬ ‫‪N‬‬
‫‪w‬‬ ‫‪S‬‬ ‫‪N‬‬
‫‪+‬‬ ‫∅‬
‫‪X‬‬
‫‪w‬‬ ‫ﻻﻳﻪ ﺧﺮﻭﺟﯽ‬
‫‪+1‬‬
‫‪w‬‬
‫‪S‬‬
‫∅‬
‫ﻻﻳﻪ ﻭﺭﻭﺩﯼ‬ ‫‪+‬‬

‫ﻻﻳﻪ ﻣﻴﺎﻧﯽ‬
‫ﺷﮑﻞ ‪ -۱‬ﻳﮏ ﺷﺒﮑﻪ ﭘﺮﺳﭙﺘﺮﻭﻥ ﭼﻨﺪ ﻻﻳﻪ ﻓﺮﺿﻲ‬

‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬

 ‫ﺩﺭ ﺍﻳﻦ ﺷﺒﮑﻪ‪ ،‬ﺑﺮﺍﻱ ﻳﮏ ﻧﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﻭﺭﻭﺩﻱ ﮐﻪ ﭘﺮﺩﺍﺯﺷـﻲ ﺭﻭﻱ ﺁﻥ‬
‫‪N o‬‬
‫‪ o (So ) Woh h (Sh ) Whp‬‬ ‫)‪(۱۲‬‬ ‫ﺻﻮﺭﺕ ﻧﻤﻲﮔﻴﺮﺩ‪ ،‬ﺑﻪﺻﻮﺭﺕ ﺯﻳﺮ ﺍﺳﺖ‬
‫‪X p‬‬

‫ﺭﺍﺑﻄﻪ ﻧﻬﺎﻳﻲ ﺣﺴﺎﺳﻴﺖ ﺧﺮﻭﺟﻲ ﻧﺴﺒﺖ ﺑﻪ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ‪ Xp‬ﺑـﺎ ﺗﻮﺟـﻪ‬ ‫‪Ni  Xi‬‬ ‫)‪(۲‬‬
‫ﺑﻪ ﺍﻳﻨﮑﻪ ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ﺑﻴﺶ ﺍﺯ ﻳﮏ ﻧﺮﻭﻥ ﻭﺟﻮﺩ ﺩﺍﺭﺩ ﺑـﻪﺻـﻮﺭﺕ ﺯﻳـﺮ‬ ‫ﻛﻪ ﺩﺭ ﺍﻳﻦ ﺭﺍﺑﻄﻪ‬
‫ﻧﻮﺷﺘﻪ ﻣﻲ ﺷﻮﺩ‬ ‫‪ i‬ﺯﻳﺮﻧﻮﻳﺲ ﻧـﺮﻭﻥ ﺩﺭ ﻻﻳـﻪ ﻭﺭﻭﺩﻱ‪ X ،‬ﻣﺘﻐﻴـﺮ ﻭﺭﻭﺩﻱ ﺑـﻪ ﺷـﺒﮑﻪ ﻭ ‪N‬‬
‫ﺳــﻴﮕﻨﺎﻝ ﺧﺮﻭﺟــﻲ ﺍﺯ ﻧــﺮﻭﻥ ﺍﺳــﺖ‪ .‬ﻧﺮﻭﻧــﻲ ﮐــﻪ ﺩﺭ ﻻﻳــﻪ ﻣﻴــﺎﻧﻲ ﺑﺎﺷــﺪ‬
‫)‪(۱۳‬‬
‫‪nh‬‬
‫‪N o‬‬
‫‪  o (So ) Woh h (S h ) Whp‬‬
‫‪X p h 1‬‬
‫ﺳﻴﮕﻨﺎﻝ ﻭﺭﻭﺩﻱ ﺭﺍ ﺑﻪ ﻓﺮﻡ ﺯﻳﺮ ﭘﺮﺩﺍﺯﺵ ﻣﻲﮐﻨﺪ‬

‫‪Sh  N p Whp   N i X hi‬‬ ‫)‪(۳‬‬

‫ﻛﻪ ﺩﺭ ﺍﻳﻦ ﺭﺍﺑﻄﻪ‬ ‫‪i p‬‬

‫ﺗﻌﺪﺍﺩ ﻧﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ﺑﺎ ‪ nh‬ﻧﺸﺎﻥ ﺩﺍﺩﻩ ﻣﻲﺷﻮﺩ‪ .‬ﺩﺭ ﻣﻄﺎﻟﻌﻪ ﺣﺎﺿﺮ‬
‫) ‪N h  h (Sh‬‬ ‫)‪(۴‬‬
‫ﻭ ﺩﺭ ﻣﺮﺣﻠــﻪ ﺁﻣــﻮﺯﺵ ﻣــﺪﻝﻫــﺎ‪ ،‬ﺗﻮﺍﺑــﻊ ﺍﻧﺘﻘــﺎﻝ ‪ Tansig‬ﻭ ‪Purelin‬‬
‫ﺑﻪﺗﺮﺗﻴﺐ ﺑﺮﺍﻱ ﻻﻳﻪﻫﺎﻱ ﻣﻴـﺎﻧﻲ ﻭ ﺧﺮﻭﺟـﻲ ﺷـﺒﮑﻪﻫـﺎ ﻣﻨﺠـﺮ ﺑـﻪ ﻧﺘـﺎﻳﺞ‬ ‫ﻛﻪ ﺩﺭ ﺍﻳﻦ ﺭﻭﺍﺑﻂ‬
‫ﻣﻨﺎﺳﺒﻲ ﺷﺪﻧﺪ ﮐﻪ ﻣﺸﺘﻖ ﻣﺮﺗﺒﻪ ﺍﻭﻝ ﺗﺎﺑﻊ ‪ Tansig‬ﺑﺎ ﺭﺍﺑﻄـﻪ ‪ ۱۴‬ﻭ ﺗـﺎﺑﻊ‬ ‫‪ S‬ﺳــﻴﮕﻨﺎﻝ ﻭﺭﻭﺩﻱ ﺑــﻪ ﻧــﺮﻭﻥ‪ ،‬ﺯﻳﺮﻧــﻮﻳﺲ ‪ p‬ﻳــﮏ ﻭﺭﻭﺩﻱ ﻣﺴــﺘﻘﻞ ﺭﺍ‬
‫‪ Purelin‬ﺑﺎ ﺭﺍﺑﻄﻪ ‪ ۱۵‬ﺑﻪﺩﺳﺖ ﻣﻲﺁﻳﻨﺪ‬ ‫ﻧﺸﺎﻥ ﻣﻲﺩﻫﺪ ﮐﻪ ﺣﺴﺎﺳﻴﺖ ﺁﻥ ﺑﻪﺩﺳﺖ ﻣﻲﺁﻳﺪ‪ Whi ،‬ﻭﺯﻥ ﺍﺗﺼﺎﻝ ﺑـﻴﻦ‬
‫ﻧﺮﻭﻥ ‪ i‬ﻭ ﻧﺮﻭﻥ ‪ h‬ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ )ﻻﻳﻪ ﺑﻌﺪﻱ(‪ ،‬ﺯﻳﺮﻧﻮﻳﺲ ‪ h‬ﺑﺮﺍﻱ ﻳـﮏ‬
‫‪‬‬ ‫‪‬‬
‫‪h (Sh )  1   2h (Sh )  1  N 2h‬‬ ‫)‪(۱۴‬‬
‫ﻧﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ‪ ‬ﻭ ﻧﺸـﺎﻥ ﺩﻫﻨـﺪﻩ ﺗـﺎﺑﻊ ﺍﻧﺘﻘـﺎﻝ ﺍﺳـﺖ‪ .‬ﺳـﻴﮕﻨﺎﻝ‬
‫‪h (So )  1‬‬ ‫)‪(۱۵‬‬ ‫ﺧﺮﻭﺟﻲ ﺍﺯ ﻧﺮﻭﻥ ‪ h‬ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ﻧﻴﺰ ﺑـﻪﺻـﻮﺭﺕ ﺯﻳـﺮ ﺑـﻪ ﻧـﺮﻭﻥ ﻻﻳـﻪ‬
‫ﺧﺮﻭﺟﻲ ﺗﻐﺬﻳﻪ ﻣﻲﺷﻮﺩ‬
‫ﺑﺎ ﺟﺎﻳﮕﺬﺍﺭﻱ ﺭﻭﺍﺑﻂ ‪ ۱۴‬ﻭ ‪ ۱۵‬ﺩﺭ ﺭﺍﺑﻄﻪ ‪ ۱۳‬ﺭﺍﺑﻄﻪ ﻧﻬـﺎﻳﻲ ﺣﺴﺎﺳـﻴﺖ‬
‫ﺑﺮﺍﻱ ﺷﺒﮑﻪ ﭘﺮﺳﭙﺘﺮﻭﻥ ﺳﻪ ﻻﻳﻪ ﻓﺮﺿﻲ ﺑﻪﺻﻮﺭﺕ ﺭﺍﺑﻄـﻪ ‪ ۱۶‬ﺑـﻪﺩﺳـﺖ‬ ‫‪So  N h Woh   N j X oj‬‬ ‫)‪(۵‬‬
‫ﻣﻲﺁﻳﺪ‬ ‫‪j h‬‬

‫) ‪N o   o (So‬‬ ‫)‪(۶‬‬
‫‪N o nh‬‬
‫‪  Woh (1  N 2h ) Whp‬‬ ‫)‪(۱۶‬‬ ‫ﻛﻪ ﺩﺭ ﺍﻳﻦ ﺭﻭﺍﺑﻂ‬
‫‪X p h 1‬‬
‫‪ j‬ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﻳﮏ ﻧﺮﻭﻥ ﺩﺭ ﻻﻳﻪ ﻣﻴﺎﻧﻲ ﻭ ﺯﻳﺮﻧﻮﻳﺲ ‪ o‬ﻣﺒـﻴﻦ ﻧـﺮﻭﻥ ﺩﺭ‬
‫ﺩﺭ ﻧﻬﺎﻳﺖ ﺍﻫﻤﻴﺖ ﻧﺴﺒﻲ ﻫﺮ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ﻧﺴﺒﺖ ﺑﻪ ﺧﺮﻭﺟﻲ ﺷـﺒﮑﻪ ﺩﺭ‬ ‫ﻻﻳﻪ ﺧﺮﻭﺟﻲ ﺍﺳﺖ‪ .‬ﺩﺭ ﻧﺘﻴﺠﻪ ﺗﻌﻴﻴﻦ ﻣﺸﺘﻖ ﻣﺮﺗﺒﻪ ﺍﻭﻝ ﺧﺮﻭﺟﻲ ﻧﺴـﺒﺖ‬
‫ﺭﻭﺵ ‪ PaD‬ﺑــﺎ ﺷــﺎﺧﺺ ﻣﺠﻤــﻮﻉ ﻣﺮﺑــﻊ ﻣﺸــﺘﻘﺎﺕ‪ ١‬ﺑــﻪﺻــﻮﺭﺕ ﺯﻳــﺮ‬ ‫ﺑﻪ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ‪ Xp‬ﺑﻪﺻﻮﺭﺕ ﺯﻳﺮ ﺍﻧﺠﺎﻡ ﻣﻲﺷﻮﺩ‬
‫ﻣﺤﺎﺳﺒﻪ ﻣﻲ ﺷﻮﺩ]‪[۲۷‬‬
‫‪Sh‬‬
‫‪ Whp‬‬ ‫)‪(۷‬‬
‫‪2‬‬ ‫‪N p‬‬
‫‪ No ‬‬
‫)‪(۱۷‬‬
‫‪NO‬‬
‫‪SSD p   ‬‬ ‫‪‬‬
‫‪‬‬ ‫‪‬‬
‫‪m 1  X p ‬‬
‫‪m‬‬
‫‪So‬‬
‫‪ Woh‬‬ ‫) ‪(۸‬‬
‫‪N h‬‬
‫ﻛﻪ ﺩﺭ ﺍﻳﻦ ﺭﺍﺑﻄﻪ‬
‫‪ m‬ﺷﻤﺎﺭﻧﺪﻩ ﺩﺍﺩﻩﻫﺎﻱ ﻣﺸﺎﻫﺪﺍﺗﻲ ﺭﻭﻱ ﺩﺍﻣﻨـﻪ ﻣﺘﻐﻴـﺮ ﻭﺭﻭﺩﻱ ‪ p‬ﺑـﺎ ‪NO‬‬ ‫‪N o N o N o N h  dN o So‬‬
‫‪‬‬ ‫‪‬‬ ‫‪‬‬
‫‪ dN h Sh ‬‬
‫‪‬‬ ‫‪‬‬ ‫)‪(۹‬‬
‫‪N p N p N h N p  dSo N h‬‬ ‫‪‬‬ ‫‪‬‬
‫‪ dSh N p ‬‬
‫ﺭﮐﻮﺭﺩ ﺛﺒﺖ ﺷﺪﻩ ﺍﺳﺖ‪ .‬ﺍﻳﻦ ﺣﺴﺎﺳﻴﺖﻫﺎ ‪  No ‬ﺑﺮﺍﻱ ﻧﻤﻮﻧـﻪ ﮔﺮﻓﺘـﻪ‬
‫‪ X ‬‬
‫‪ p‬‬ ‫ﺍﺯﻃﺮﻓﻲ ﺍﺯ ﺭﻭﺍﺑـﻂ ‪ ۴‬ﻭ ‪ ۶‬ﺭﻭﺍﺑـﻂ ‪ ۱۰‬ﻭ ‪ ۱۱‬ﺣﺎﺻـﻞ ﺷـﺪﻩ ﻭ ﻫﻤﭽﻨـﻴﻦ‬
‫ﺷـﺪﻩ ﺍﺯ ﺑﺎﻧــﮏ ﺍﻃﻼﻋـﺎﺗﻲ ﻣــﺪﻝ )ﮐﻠﻴـﻪ ﺭﮐﻮﺭﺩﻫــﺎﻱ ﺛﺒـﺖ ﺷــﺪﻩ( ﻭ ﺩﺭ‬ ‫ﺭﺍﺑﻄﻪ ‪ ۹‬ﺑﻪ ﻓﺮﻡ ‪ ۱۲‬ﺑﺎﺯﻧﻮﻳﺴﻲ ﻣﻲﺷﻮﺩ‬
‫ﻓﻀﺎﻱ ‪ n‬ﺑﻌﺪﻱ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﻣﺤﺎﺳﺒﻪ ﻣﻲﺷﻮﻧﺪ‪ .‬ﻳـﮏ ﺣﺴﺎﺳـﻴﺖ‬
‫ﻣﺜﺒﺖ ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﺭﺍﺑﻄﻪ ﻣﺴـﺘﻘﻴﻢ ﻭ ﻳـﮏ ﻣﻘـﺪﺍﺭ ﻣﻨﻔـﻲ ﻧﺸـﺎﻥ ﺩﻫﻨـﺪﻩ‬ ‫‪dN o‬‬
‫) ‪ o (So‬‬ ‫)‪(۱۰‬‬
‫ﺭﺍﺑﻄﻪ ﻣﻌﮑﻮﺱ ﻣﺘﻐﻴﺮ ﻣﻮﺭﺩ ﻧﻈﺮ ﺑﺎ ﺧﺮﻭﺟﻲ ﺍﺳـﺖ‪ .‬ﺑـﺮﺍﻱ ﺍﻳـﻦ ﺭﻭﺍﺑـﻂ‬ ‫‪dSo‬‬

‫‪1‬‬
‫‪dN h‬‬
‫) ‪ h (Sh‬‬ ‫)‪(۱۱‬‬
‫)‪Sum of Square Derivatives (SSD‬‬ ‫‪dS h‬‬

‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫‪ ‬‬

‫ﻧﻬـﺎﻳﻲ ﺭﻭﺷـﻬﺎﻱ ﻣــﺬﮐﻮﺭ‪ ،‬ﻓﺮﺿـﻴﺎﺗﻲ ﺩﺭ ﻧﻈــﺮ ﮔﺮﻓﺘـﻪ ﻣــﻲﺷـﻮﺩ‪ .‬ﭼﻨــﻴﻦ‬ ‫ﺍﮔﺮ ﺩﺍﺩﻩﻫﺎ ﺑـﻪﺻـﻮﺭﺕ ﺧـﺎﻡ ﻭﺍﺭﺩ ﺷـﺒﮑﻪ ﺷـﻮﻧﺪ ﻧﺘﻴﺠـﻪ ﺣﺎﺻـﻞ ﺑﻴـﺎﻧﮕﺮ‬
‫ﺭﻭﺷﻬﺎﻳﻲ ﮐﻪ ﺩﺭ ﺍﺑﺘﺪﺍ ﺑﻪﻋﻠﺖ ﺩﺭ ﻧﻈﺮ ﮔﺮﻓﺘﻦ ﻋﻮﺍﻣﻞ ﻣﺨﺘﻠﻒ ﻣﻲﺗﻮﺍﻧﻨـﺪ‬ ‫ﺣﺴﺎﺳﻴﺖ ﻣﻄﻠﻖ ﺍﺳﺖ ﮐﻪ ﺑﺎﻳﺪ ﺍﺻﻼﺡ ﺷﻮﺩ‪ .‬ﺍﻣﺎ ﺍﮔﺮ ﺩﺍﺩﻩﻫـﺎﻱ ﻭﺭﻭﺩﻱ‬
‫ﺟﺎﻣﻌﻴﺖ ﺩﺍﺷﺘﻪ ﺑﺎﺷـﻨﺪ‪ ،‬ﺑـﺎ ﺍﻋﻤـﺎﻝ ﻓﺮﺿـﻴﺎﺕ ﺳـﺎﺩﻩ ﮐﻨﻨـﺪﻩ‪ ،‬ﺧﻄـﺎﻫـﺎﻱ‬ ‫ﺑﻪ ﺷﺒﮑﻪ ﻧﺮﻣﺎﻟﻴﺰﻩ ﺷﻮﻧﺪ ﺭﻭﺍﺑﻂ ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ ﺑﻴـﺎﻧﮕﺮ ﺣﺴﺎﺳـﻴﺖ ﻧﺴـﺒﻲ‬
‫ﺑﺰﺭﮔﻲ ﺩﺭ ﭘﺎﺳﺦ ﻧﻬﺎﻳﻲ ﺁﻧﻬﺎ ﺍﻳﺠﺎﺩ ﻣﻲﺷﻮﺩ ﮐﻪ ﺩﻗﺖ ﺁﻧﻬـﺎ ﺭﺍ ﺩﺭ ﻣﻘﺎﻳﺴـﻪ‬ ‫ﺧﻮﺍﻫﻨــﺪ ﺑــﻮﺩ]‪ .[۴‬ﺷــﺎﺧﺺ ‪ SSD‬ﺍﻫﻤﻴــﺖ ﻧﺴــﺒﻲ ﺷــﺮﻛﺖ ﻫــﺮ ﻣﺘﻐﻴــﺮ‬
‫ﺑﺎ ﻣﺪﻝﻫﺎﻱ ﺟﻌﺒﻪ ﺳﻴﺎﻩ ﮐﻪ ﺑـﺮﺍﻱ ﻫـﺮ ﻣـﻮﺭﺩ ﻣﻄﺎﻟﻌـﺎﺗﻲ ﺧـﺎﺹ ﮐـﺎﻟﻴﺒﺮﻩ‬ ‫ﻭﺭﻭﺩﻱ ﺩﺭ ﻣﺤﺎﺳﺒﻪ ﻣﺘﻐﻴﺮ ﺧﺮﻭﺟـﻲ ﺷـﺒﻜﻪ ﺭﺍ ﺑـﻪﺩﺳـﺖ ﻣـﻲﺁﻭﺭﺩ ﻭﻟـﻲ‬
‫ﻣﻲﺷﻮﻧﺪ ﭘﺎﻳﻴﻦ ﻣـﻲﺁﻭﺭﺩ‪ .‬ﻳﮑـﻲ ﺍﺯ ﺩﻻﻳـﻞ ﺑﺮﺗـﺮﻱ ﭘﺎﺳـﺦ ﺭﻭﺵ ﭘـﻨﻤﻦ‬ ‫ﺟﻬﺖ ﺗﺄﺛﻴﺮ ﺭﺍ ﻣﺸﺨﺺ ﻧﻤﻲﮐﻨﺪ‪ .‬ﻫﺮ ﺍﻧﺪﺍﺯﻩ ﻣﻘﺪﺍﺭ ﺍﻳﻦ ﺷـﺎﺧﺺ ﺑﻴﺸـﺘﺮ‬
‫ﻧﺴﺒﺖ ﺑﻪ ﺩﻭ ﺭﻭﺵ ﺗﺠﺮﺑﻲ ﺩﻳﮕﺮ ﺑـﻪﮐـﺎﺭ ﺭﻓﺘـﻪ‪ ،‬ﺗـﺄﺛﻴﺮ ﮐﻤﺘـﺮ ﻓﺮﺿـﻴﺎﺕ‬ ‫ﺑﺎﺷﺪ‪ ،‬ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﺗﺄﺛﻴﺮ ﺑﻴﺸﺘﺮ ﻣﺘﻐﻴﺮ ﻣﻮﺭﺩ ﻧﻈﺮ ﺭﻭﻱ ﺧﺮﻭﺟﻲ ﺍﺯ ﻣـﺪﻝ‬
‫ﺳﺎﺩﻩ ﮐﻨﻨﺪﻩ ﺑـﺮ ﺭﻭﺵ ﻣـﺬﮐﻮﺭ ﺍﺳـﺖ‪ .‬ﺑـﻪﻣﻨﻈـﻮﺭ ﻣـﺪﻝﺳـﺎﺯﻱ ﺑـﺎ ﺭﻭﺵ‬ ‫ﺍﺳﺖ‪ .‬ﺑﺮﺍﻱ ﻣﺸﺨﺺ ﮐﺮﺩﻥ ﻣﻴﺰﺍﻥ ﭘﺮﺍﮐﻨﺪﮔﻲ ﻭ ﺑﻪﻣﻨﻈﻮﺭ ﺗـﺄﺛﻴﺮ ﻣﻘـﺎﺩﻳﺮ‬
‫‪ ، MLR‬ﺩﺍﺩﻩﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺑﻪﺻﻮﺭﺕ ﻏﻴﺮﻧﺮﻣـﺎﻝ ﺍﺳـﺘﻔﺎﺩﻩ ﺷـﺪ ﻭ ﭘﺎﺳـﺦ‬ ‫ﻣﺨﺘﻠﻒ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﻣﺘﻐﻴﺮ ﺧﺮﻭﺟﻲ ﻧﺴﺒﺖ ﺑﻪ ﻳﮏ ﻣﺘﻐﻴـﺮ ﻭﺭﻭﺩﻱ‬
‫ﻣﺪﻝ ﺣﺎﻭﻱ ﻣﻘﺎﺩﻳﺮ ﺳﻄﻮﺡ ﻣﻌﻨﻲﺩﺍﺭﻱ ﻧﺰﺩﻳﮏ ﺑـﻪ ﺻـﻔﺮ ﺑـﺮﺍﻱ ﺗﻤـﺎﻣﻲ‬ ‫ﺩﺭ ﮐﻞ ﻓﻀﺎﻱ ﻭﺭﻭﺩﻱ ﺍﺯ ﺭﻭﺵ ﺁﻧﺎﻟﻴﺰ ﺁﻣﺎﺭﻱ ﻣﻘـﺎﺩﻳﺮ ﺣﺴﺎﺳـﻴﺖﻫـﺎﻱ‬
‫ﻭﺭﻭﺩﻱﻫﺎ ﺑﻮﺩ ﮐﻪ ﺑﻴﺎﻧﮕﺮ ﻣﺪﻝﺳﺎﺯﻱ ﺧﻮﺏ ﺍﺳﺖ‪ .‬ﺍﮔﺮ ﺍﻳﻦ ﺳﻄﻮﺡ ﺑﺮﺍﻱ‬ ‫ﺑﻪ ﺩﺳﺖ ﺁﻣﺪﻩ ﺍﺳـﺘﻔﺎﺩﻩ ﻣـﻲﺷـﻮﺩ]‪ .[۴‬ﺩﺭ ﺍﻳـﻦ ﺭﻭﺵ ‪ ۵‬ﺩﺭﺻـﺪ ﺁﻣـﺎﺭﻱ‬
‫ﻭﺭﻭﺩﻱ ﺧﺎﺻﻲ ﺩﺍﺭﺍﻱ ﻣﻘﺎﺩﻳﺮ ﺑـﺎﻻ ﺑﺎﺷـﺪ ﺑـﻪ ﺍﻳـﻦ ﻣﻌﻨـﻲ ﺍﺳـﺖ ﮐـﻪ ﺩﺭ‬ ‫)‪ ۱۰ D‬ﺩﺭﺻﺪ‪ ۲۵ D ،‬ﺩﺭﺻﺪ‪ ۵۰ D ،‬ﺩﺭﺻـﺪ‪ ۷۵ D ،‬ﺩﺭﺻـﺪ ﻭ ‪۹۰ D‬‬
‫ﻓﺮﺍﻳﻨﺪ ﻣﺪﻝﺳﺎﺯﻱ ﻻﺯﻡ ﺍﺳـﺖ ﮐـﻪ ﺁﻥ ﻭﺭﻭﺩﻱ ﺣـﺬﻑ ﺷـﻮﺩ‪ ،‬ﻫﻤﭽﻨـﻴﻦ‬ ‫ﺩﺭﺻﺪ( ﻣﻘﺎﺩﻳﺮﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﺧﺮﻭﺟـﻲ ﻧﺴـﺒﺖ ﺑـﻪ ﻭﺭﻭﺩﻱ ﻣﺮﺑﻮﻃـﻪ‬
‫ﻧﺘﺎﻳﺞ ﺭﻭﺵ ﺭﮔﺮﺳﻴﻮﻧﻲ ﺣﺎﻭﻱ ﺑﺮﺁﻭﺭﺩ ﻣﻘﺎﺩﻳﺮ ﻣﻨﻔﻲ ﺑﺮﺍﻱ ﺗﺒﺨﻴﺮ ﺍﺳـﺖ‬ ‫ﻣﺤﺎﺳﺒﻪ ﻣﻲﺷﻮﻧﺪ ﮐﻪ ﺗﻮﺿﻴﺢ ﺍﻳﻦ ﺷﺎﺧﺼﻬﺎ ﺑﻪ ﺍﻳﻦ ﺻﻮﺭﺕ ﺍﺳـﺖ ﮐـﻪ ‪D‬‬
‫)ﺑــﻪﺻــﻮﺭﺕ ﺟﺰﺋــﻲ(‪ .‬ﺍﻳــﻦ ﺧﻄــﺎ ﻧﺎﺷــﻲ ﺍﺯ ﻧﺤــﻮﻩ ﻣــﺪﻝﺳــﺎﺯﻱ ﺭﻭﺵ‬ ‫‪ ۱۰‬ﺩﺭﺻﺪ ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﻣﻘﺪﺍﺭﻱ ﺑﺮﺍﻱ ﺣﺴﺎﺳﻴﺖ ﻧﺴـﺒﻲ ﺍﺳـﺖ ﮐـﻪ ‪۹۰‬‬
‫ﺭﮔﺮﺳﻴﻮﻥ ﺧﻄﻲ ﭼﻨﺪ ﻣﺘﻐﻴـﺮﻩ ﺍﺳـﺖ ﮐـﻪ ﻣﻮﻗﻌﻴـﺖ ﺑﻬﺘـﺮﻳﻦ ﺍﺑـﺮ ﺻـﻔﺤﻪ‬ ‫ﺩﺭﺻﺪ ﻣﻘﺎﺩﻳﺮ ﺍﺯ ﺁﻥ ﺑﻴﺸﺘﺮ ﻭ ‪ ۱۰‬ﺩﺭﺻﺪ ﺍﺯ ﺁﻥ ﮐﻤﺘﺮ ﻫﺴـﺘﻨﺪ‪ .‬ﺑﻨـﺎﺑﺮﺍﻳﻦ‬
‫ﺭﮔﺮﺳﻴﻮﻥ ﺑﺎ ﺍﺳﺘﻔﺎﺩﻩ ﺍﺯ ﮐﻤﺘﺮﻳﻦ ﺧﻄﺎﻱ ﻣﺮﺑﻌـﺎﺕ ﺭﺍ ﮐـﻪ ﻣـﻲﺗـﻮﺍﻥ ﺑـﻴﻦ‬ ‫ﺍﮔﺮ ﻣﺜﺒﺖ ﺷﻮﺩ ﻧﺸﺎﻥ ﻣﻲﺩﻫﺪ ﮐﻪ ﺍﺣﺘﻤﺎﻝ ﺍﻳﻨﮑﻪ ﻣﻘﺪﺍﺭ ﺣﺴﺎﺳﻴﺖ ﻧﺴـﺒﻲ‬
‫ﻣﺸﺎﻫﺪﺍﺕ ﺑﺮﺍﺯﺵ ﺩﺍﺩ‪ ،‬ﺗﻌﻴﻴﻦ ﻣﻲﻛﻨﺪ‪ .‬ﺩﺭ ﺍﺭﺗﺒﺎﻁ ﺑﺎ ﻣﺪﻝﺳﺎﺯﻱ ﺗﻮﺳـﻂ‬ ‫ﻣﺜﺒﺖ ﺑﺎﺷﺪ ﺑﺎﻻﻱ ‪ ۹۰‬ﺩﺭﺻﺪ ﺍﺳﺖ ﻳﺎ ﺑﻪﻋﺒﺎﺭﺗﻲ ﺍﺣﺘﻤﺎﻝ ﺍﻳﻨﮑﻪ ﺧﺮﻭﺟﻲ‬
‫‪ ، ANN‬ﺟــﺪﻭﻝ ‪ ۲‬ﺗﻌــﺪﺍﺩﻱ ﺍﺯ ﺳــﺎﺧﺘﺎﺭﻫﺎﻱ ﺑﻬﻴﻨــﻪ ﺑــﻪﺩﺳــﺖ ﺁﻣــﺪﻩ ﺍﺯ‬ ‫ﺑﺎ ﺍﻓﺰﺍﻳﺶ ﻭﺭﻭﺩﻱ ﻣﻮﺭﺩ ﻧﻈﺮ ﺍﻓﺰﺍﻳﺶ ﻳﺎﺑﺪ ﺑﺎﻻﻱ ‪ ۹۰‬ﺩﺭﺻﺪ ﺍﺳـﺖ‪D .‬‬
‫ﻣـﺪﻝﻫـﺎﻱ ‪ ANN‬ﺭﺍ ﻧﺸــﺎﻥ ﻣـﻲﺩﻫـﺪ‪ .‬ﺑــﺮﺍﻱ ﻣﻘﺎﻳﺴـﻪ ﻧﺘــﺎﻳﺞ ﺩﻭ ﺭﻭﺵ‬ ‫‪ ۹۰‬ﺩﺭﺻﺪ ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﻣﻘﺪﺍﺭﻱ ﺑﺮﺍﻱ ﺣﺴﺎﺳﻴﺖ ﻧﺴـﺒﻲ ﺍﺳـﺖ ﮐـﻪ ‪۹۰‬‬
‫ﺭﮔﺮﺳﻴﻮﻧﻲ ﻭ ‪ ، ANN‬ﻧﺘﺎﻳﺞ ﻣﺪﻝﻫﺎ ﺑﻪﻋﻨﻮﺍﻥ ﻣﺜﺎﻝ ﺑـﺮﺍﻱ ﺷـﻬﺮ ﺍﺭﻭﻣﻴـﻪ‬ ‫ﺩﺭﺻﺪ ﻣﻘﺎﺩﻳﺮ ﺍﺯ ﺁﻥ ﮐﻤﺘﺮ ﺍﺳﺖ‪ ،‬ﺑﻨﺎﺑﺮﺍﻳﻦ ﺍﮔﺮ ﻣﻨﻔﻲ ﺑﺎﺷﺪ ﻧﺸﺎﻥ ﻣﻲﺩﻫﺪ‬
‫ﺑﻪﺗﺮﺗﻴﺐ ﺩﺭ ﺷﻜﻠﻬﺎﻱ ‪ ۲‬ﻭ ‪ ۳‬ﺭﺳﻢ ﺷﺪﻩﺍﻧﺪ‪.‬‬ ‫ﮐﻪ ﺑﻪ ﺍﺣﺘﻤﺎﻝ ﺑﺎﻻﻱ ‪ ۹۰‬ﺩﺭﺻﺪ‪ ،‬ﺧﺮﻭﺟﻲ ﺑﺎ ﺍﻓﺰﺍﻳﺶ ﻭﺭﻭﺩﻱ ﻣﻮﺭﺩ ﻧﻈﺮ‬
‫ﺟﺪﻭﻝ ‪ ۳‬ﺷﺎﻣﻞ ‪ R2‬ﻭ ‪ RMSE‬ﺑﻪﺩﺳﺖ ﺁﻣـﺪﻩ ﺍﺯ ﺗﻤـﺎﻣﻲ ﺭﻭﺷـﻬﺎﻱ‬ ‫ﮐﺎﻫﺶ ﻣﻲﻳﺎﺑﺪ‪ .‬ﺗﻮﺿﻴﺢ ‪ ۲۵ D‬ﺩﺭﺻﺪ ‪ ۵۰ D ،‬ﺩﺭﺻﺪ ﻭ ‪ ۷۵ D‬ﺩﺭﺻـﺪ‬
‫ﺑﻪﮐﺎﺭ ﮔﺮﻓﺘﻪ ﺷﺪﻩ ﺑﺮﺍﻱ ﻫﺮ ﻣﻨﻄﻘﻪ ﺍﺳﺖ‪ .‬ﺍﺯ ﺟﺪﻭﻝ ‪ ۳‬ﺍﺳﺘﻨﺒﺎﻁ ﻣـﻲﺷـﻮﺩ‬ ‫ﻣﺎﻧﻨــﺪ ‪ ۱۰ D‬ﺩﺭﺻــﺪ ﻭ ‪ ۹۰ D‬ﺩﺭﺻــﺪ ﺍﺳــﺖ‪ .‬ﻭﺭﻭﺩﻱ ﮐــﻪ ﻣﻘــﺎﺩﻳﺮ‬
‫ﮐﻪ ﻣﺪﻝﺳﺎﺯﻱ ﺑﺎ ‪ ANNs‬ﺩﺭ ﻫﺮ ﺩﻭ ﻣﻨﻄﻘﻪ ﻣﻴـﺰﺍﻥ ﺗﺒﺨﻴـﺮ ﺭﻭﺯﺍﻧـﻪ ﺭﺍ ﺩﺭ‬ ‫ﺩﺭﺻﺪﻫﺎﻱ ﺣﺴﺎﺳﻴﺖ ﻧﺴـﺒﻲ ﺁﻥ ﺩﺭ ﺍﻃـﺮﺍﻑ ﺧـﻂ ﭘﺎﻳـﻪ ﻣﺘﻤﺮﮐـﺰ ﺷـﺪﻩ‬
‫ﻣﻘﺎﻳﺴﻪ ﺑﺎ ﺩﻳﮕﺮ ﺭﻭﺷﻬﺎ ﺑﺎ ﻣﻴﺰﺍﻥ ﺧﻄﺎﻱ ﭘﺎﺋﻴﻦﺗـﺮﻱ ﺗﺨﻤـﻴﻦ ﺯﺩﻩ ﺍﺳـﺖ‪.‬‬ ‫ﺑﺎﺷﺪ‪ ،‬ﺩﺍﺭﺍﻱ ﺍﺛﺮ ﮐﻤﺘﺮ ﺑﺮ ﺭﻭﻱ ﺧﺮﻭﺟـﻲ ﻧﺴـﺒﺖ ﺑـﻪ ﻭﺭﻭﺩﻱ ﺍﺳـﺖ ﮐـﻪ‬
‫ﺍﺯ ﺩﻻﻳﻞ ﺍﻳـﻦ ﺑﺮﺗـﺮﻱ ﻣـﻲ ﺗـﻮﺍﻥ ﺑـﻪ ﺍﻧﻌﻄـﺎﻑﭘـﺬﻳﺮ ﺑـﻮﺩﻥ ‪ ANNs‬ﺩﺭ‬ ‫ﺩﺳﺘﻪ ﺩﺭﺻﺪﻫﺎﻱ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﺁﻥ ﺩﺍﺭﺍﻱ ﭘـﺮﺍﮐﻨﺶ ﻭﺳـﻴﻊﺗـﺮﻱ ﺩﺭ‬
‫ﺗﺨﻤﻴﻦ ﺭﻭﻧﺪﻫﺎﻱ ﺧﻄﻲ ﻭ ﻏﻴﺮﺧﻄﻲ ﺍﺷﺎﺭﻩ ﮐﺮﺩ‪ .‬ﺑـﻪﻃـﻮﺭ ﻣﺸـﺎﺑﻪ ﺩﻳﮕـﺮ‬ ‫ﺍﻃﺮﺍﻑ ﺧﻂ ﭘﺎﻳﻪ ﺍﺳﺖ‪.‬‬
‫ﻣﻄﺎﻟﻌﺎﺕ ﺻﻮﺭﺕ ﮔﺮﻓﺘﻪ ﺩﺭ ﺯﻣﻴﻨﻪ ﺑﺮﺁﻭﺭﺩ ﺗﺒﺨﻴﺮ ﺗﻮﺳﻂ ﻣﺤﻘﻘﺎﻧﻲ ﻧﻈﻴﺮ‬
‫ﻣﻘﺪﻡﻧﻴـﺎ ﻭ ﻫﻤﮑـﺎﺭﺍﻥ‪ ،‬ﺭﺣﻴﻤـﻲﺧـﻮﺏ‪ ،‬ﺗﺮﺍﺋـﻮﺭﻩ ﻭ ﻫﻤﮑـﺎﺭﺍﻥ‪ ، ١‬ﻋﻤـﺪﺗﺎﹰ‬ ‫‪ -۵‬ﻧﺘﺎﻳﺞ ﻭ ﺑﺤﺚ‬
‫ﺑﺮﺗﺮﻱ ﭘﺎﺳﺦ ‪ ANNs‬ﺭﺍ ﻧﺴﺒﺖ ﺑﻪ ﺭﻭﺷﻬﺎﻱ ﺗﺠﺮﺑﻲ ﻭ ﻣﻄﺎﻟﻌـﺎﺕ ﺍﻧﺠـﺎﻡ‬ ‫‪ -۱-۵‬ﻧﺘﺎﻳﺞ ﻣﺪﻝﺳﺎﺯﻱ ﺗﺒﺨﻴﺮ‬
‫ﺷﺪﻩ ﺗﻮﺳﻂ ﻃﺒﺮﻱ‪ ،‬ﺩﻳﺴﻮﺍﻝ ﻭ ﭘﺎﻝ ﺩﻗﻴﻖﺗﺮ ﺑﻮﺩﻥ ‪ ANNs‬ﺭﺍ ﻧﺴﺒﺖ ﺑـﻪ‬ ‫ﺑﺮﺍﻱ ﻫﺮ ﺩﻭ ﻣﻨﻄﻘﻪ ﻣﻄﺎﻟﻌﺎﺗﻲ ﻧﺘﺎﻳﺞ ﺑـﻪﮐـﺎﺭﮔﻴﺮﻱ ﺳـﻪ ﺭﻭﺵ ﮐﻼﺳـﻴﮏ‬
‫ﻣﺪﻝ ‪ MLR‬ﮔﺰﺍﺭﺵ ﮐﺮﺩﻩﺍﻧﺪ]‪ ۱۳-۱۰‬ﻭ ‪ .[۱۵‬ﻫﻤﺎﻥﻃﻮﺭﻱ ﮐـﻪ ﻧﺘـﺎﻳﺞ‬ ‫)ﺗﻌــﺎﺩﻝ ﺍﻧــﺮﮊﻱ‪ ،‬ﺁﻳﺮﻭﺩﻳﻨــﺎﻣﻴﮑﻲ ﻭ ﭘــﻨﻤﻦ( ﺑــﺎ ﺗﻮﺟــﻪ ﺑــﻪ ﺷــﺎﺧﺺ ‪R2‬‬
‫ﺟﺪﻭﻝ ‪ ۳‬ﻧﻴﺰ ﺣﺎﮐﻲ ﺍﺯ ﻋﻤﻠﮑﺮﺩ ﺿﻌﻴﻒﺗﺮ ﺭﻭﺵ ﺭﮔﺮﺳﻴﻮﻧﻲ ﻧﺴـﺒﺖ ﺑـﻪ‬ ‫ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ‪ ،‬ﻧﺸﺎﻥ ﺩﻫﻨﺪﻩ ﺑﺮﺗﺮ ﺑﻮﺩﻥ ﺭﻭﺵ ﺗﺮﻛﻴﺒﻲ ﭘـﻨﻤﻦ ﻧﺴـﺒﺖ ﺑـﻪ‬
‫ﻣﺪﻝ ‪ ANN‬ﺩﺭ ﻣﺪﻝﺳﺎﺯﻱ ﻓﺮﺍﻳﻨﺪ ﻏﻴﺮﺧﻄﻲ ﺗﺒﺨﻴﺮ ﺍﺳـﺖ‪ .‬ﻫﻤﭽﻨـﻴﻦ ﺍﺯ‬ ‫ﺩﻭ ﺭﻭﺵ ﺩﻳﮕــﺮ ﺍﺳــﺖ‪ .‬ﺑﻌــﺪ ﺍﺯ ﺭﻭﺵ ﭘــﻨﻤﻦ ﺩﺭ ﻫــﺮ ﺩﻭ ﻣﻨﻄﻘــﻪ ﺭﻭﺵ‬
‫ﺟﺪﻭﻝ ‪ ۳‬ﻧﺘﻴﺠﻪ ﻣﻲﺷﻮﺩ ﮐﻪ ﺭﻭﺷﻬﺎﻱ ﺟﻌﺒـﻪ ﺳـﻴﺎﻩ ﺭﮔﺮﺳـﻴﻮﻧﻲ ﻭ ‪ANN‬‬ ‫ﺁﻳﺮﻭﺩﻳﻨﺎﻣﻴﻜﻲ ﺑﻬﺘﺮ ﺍﺯ ﺗﻌﺎﺩﻝ ﺍﻧﺮﮊﻱ ﺭﻓﺘﺎﺭ ﻛﺮﺩ‪ .‬ﺑﻪﻣﻨﻈﻮﺭ ﻣﻘﺎﻳﺴﻪ ﮐﻠﻴـﻪ‬
‫ﮐﻪ ﺑﺮﺍﻱ ﻣـﻮﺭﺩ ﻣﻄﺎﻟﻌـﺎﺗﻲ ﺗﺒﺨﻴـﺮ ﺭﻭﺯﺍﻧـﻪ ﻭ ﺑـﺮﺍﻱ ﺩﻭ ﻣﻨﻄﻘـﻪ ﺗﺒﺮﻳـﺰ ﻭ‬ ‫ﻣﺪﻝﻫﺎ ﻧﺘﺎﻳﺞ ﺍﻳﻦ ﻗﺴﻤﺖ ﺑـﺮﺍﻱ ﺩﻭ ﮔـﺮﻭﻩ ﺁﻣﻮﺯﺷـﻲ ﻭ ﺁﺯﻣﺎﻳﺸـﻲ ﺟـﺪﺍ‬
‫ﺍﺭﻭﻣﻴﻪ ﮐﺎﻟﻴﺒﺮﻩ ﺷﺪﻩﺍﻧﺪ‪ ،‬ﭘﺎﺳﺨﻬﺎﻳﻲ ﺑﻬﺘﺮ ﻭ ﺑـﺎ ﺧﻄـﺎﻱ ﭘـﺎﻳﻴﻦ ﻧﺴـﺒﺖ ﺑـﻪ‬ ‫ﺷﺪ‪ .‬ﺭﻭﺷﻬﺎﻱ ﺗﺠﺮﺑﻲ ﻣﺒﺘﻨـﻲ ﺑـﺮ ﺭﻭﺍﺑـﻂ ﻓﻴﺰﻳﮑـﻲ ﻫﺴـﺘﻨﺪ ﮐـﻪ ﻋﻮﺍﻣـﻞ‬
‫‪1‬‬
‫ﻣﺨﺘﻠﻔﻲ ﺩﺭ ﺑﻪﺩﺳﺖ ﺁﻣﺪﻥ ﺁﻧﻬﺎ ﺩﺧﺎﻟﺖ ﺩﺍﺭﻧـﺪ‪ .‬ﺑـﻪﻋﻠـﺖ ﭘﻴﭽﻴـﺪﻩ ﺑـﻮﺩﻥ‬
‫‪Traore et al.‬‬
‫ﺍﻧﺪﺍﺯﻩﮔﻴﺮﻱ ﭘﺎﺭﻩﺍﻱ ﺍﺯ ﻋﻮﺍﻣﻞ ﺗﺄﺛﻴﺮﮔﺬﺍﺭ ﻭ ﺩﺭ ﺭﻭﻧﺪ ﺳﺎﺩﻩﺳﺎﺯﻱ ﺭﻭﺍﺑـﻂ‬

‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬

 ‫ﺩﺭ ﺯﻣﺎﻥ ﻗﺒﻞ )‪ (t-1‬ﺩﺍﺭﺩ‪ .‬ﺑﻨﺎﺑﺮﺍﻳﻦ ﻳﮑـﻲ ﺍﺯ ﺩﻻﻳـﻞ ﺑﺮﺗـﺮﻱ ﻣـﺪﻝﻫـﺎﻱ‬ ‫ﺭﻭﺷﻬﺎﻱ ﺗﺠﺮﺑﻲ ﮐﻪ ﺑﺮ ﺍﺳﺎﺱ ﺭﻭﺍﺑﻂ ﻓﻴﺰﻳﮑﻲ ﻫﺴﺘﻨﺪ‪ ،‬ﺗﻮﻟﻴـﺪ ﮐـﺮﺩﻩﺍﻧـﺪ‪.‬‬
‫ﺭﮔﺮﺳﻴﻮﻧﻲ ﻭ‪ ANN‬ﺑﻪ ﺭﻭﺷﻬﺎﻱ ﺗﺠﺮﺑﻲ ﺑﻪﮐﺎﺭ ﺭﻓﺘﻪ ﺭﺍ ﻣﻲﺗﻮﺍﻥ ﺩﺭ ﻧﻈـﺮ‬ ‫ﺍﺯ ﻃــﺮﻑ ﺩﻳﮕــﺮ ﻣﺎﻧﻨــﺪ ﺍﻏﻠــﺐ ﭘﺪﻳــﺪﻩﻫــﺎﻱ ﻫﻴــﺪﺭﻭﻟﻮﮊﻳﮑﻲ‪ ،‬ﺗﺒﺨﻴــﺮ ﺭﺍ‬
‫ﮔﺮﻓﺘﻦ ﺗﺒﺨﻴﺮ ﺑﻪﻋﻨﻮﺍﻥ ﻳﮏ ﻓﺮﺍﻳﻨﺪ ﻣﺎﺭﮐﻮﻑ ﺩﺍﻧﺴﺖ ﮐﻪ ﺑـﺎ ﻭﺍﺭﺩ ﮐـﺮﺩﻥ‬ ‫ﻣﻲﺗﻮﺍﻥ ﻳﮏ ﻓﺮﺍﻳﻨﺪ ﻣﺎﺭﮐﻮﻑ‪ ١‬ﺩﺭ ﻧﻈﺮ ﮔﺮﻓـﺖ‪ .‬ﺩﺭ ﭼﻨـﻴﻦ ﻓﺮﺍﻳﻨـﺪﻫﺎﻳﻲ‬
‫ﻣﺆﻟﻔــﻪ ﺗﺒﺨﻴــﺮ ﺭﻭﺯ ﮔﺬﺷــﺘﻪ ‪ Et-1‬ﺩﺭ ﻣــﺪﻝﺳــﺎﺯﻱ‪ ،‬ﺩﻗــﺖ ﻣــﺪﻝ ﺩﺭ ﺁﻥ‬ ‫ﺍﺣﺘﻤﺎﻝ ﻭﻗﻮﻉ ﻳﮏ ﺣﺎﻟﺖ ﺍﻗﻠﻴﻤﻲ ﺩﺭ ﺯﻣـﺎﻥ ‪ t‬ﺑﺴـﺘﮕﻲ ﺑـﻪ ﻭﺿـﻌﻴﺖ ﺁﻥ‬
‫ﺍﻓﺰﺍﻳﺶ ﻣﻲﻳﺎﺑﺪ‪.‬‬
‫‪1‬‬
‫‪Markovian Process‬‬

‫ﺟﺪﻭﻝ ‪ -۲‬ﺗﺮﮐﻴﺒﺎﺕ ﺑﻬﻴﻨﻪ ﻣﺪﻝﺳﺎﺯﻱ ﺗﺒﺨﻴﺮ ﺑﺎ ‪ ANNs‬ﺑﺮﺍﻱ ﺷﻬﺮﻫﺎﻱ ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ‬

‫ﺍﺭﻭﻣﻴﻪ‬ ‫ﺗﺒﺮﻳﺰ‬
‫‪Epoch‬‬ ‫ﺳﺎﺧﺘﺎﺭ ﻻﻳﻪﻫﺎ‬
‫ﺁﺯﻣﺎﻳﺶ‬ ‫ﺁﻣﻮﺯﺵ‬ ‫ﺁﺯﻣﺎﻳﺶ‬ ‫ﺁﻣﻮﺯﺵ‬ ‫ﺗﺮﮐﻴﺒﺎﺕ‬
‫‪RMSE‬‬ ‫‪RMSE‬‬ ‫‪RMSE‬‬ ‫‪RMSE‬‬ ‫ﺍﺭﻭﻣﻴﻪ‬ ‫ﺗﺒﺮﻳﺰ‬ ‫ﺍﺭﻭﻣﻴﻪ‬ ‫ﺗﺒﺮﻳﺰ‬
‫)‪(mm/d‬‬ ‫‪R2‬‬ ‫)‪(mm/d‬‬ ‫‪R2‬‬ ‫)‪(mm/d‬‬ ‫‪R2‬‬ ‫)‪(mm/d‬‬ ‫‪R2‬‬

‫‪0/109‬‬ ‫‪0/828‬‬ ‫‪0/079‬‬ ‫‪0/889‬‬ ‫‪0/077‬‬ ‫‪0/893‬‬ ‫‪0/078‬‬ ‫‪0/888‬‬ ‫‪50‬‬ ‫‪80‬‬ ‫‪6-6-1‬‬ ‫‪6-8-1‬‬ ‫‪Et-1 , RH , Ra ,‬‬
‫‪T, P, U2‬‬
‫‪-‬‬ ‫‪-‬‬ ‫‪-‬‬ ‫‪-‬‬ ‫‪0/078‬‬ ‫‪0/891‬‬ ‫‪0/078‬‬ ‫‪0/886‬‬ ‫‪-‬‬ ‫‪70‬‬ ‫‪-‬‬ ‫‪4-10-1‬‬ ‫‪Et-1 , Ra , T,‬‬
‫‪U2‬‬
‫‪0/107‬‬ ‫‪0/834‬‬ ‫‪0/082‬‬ ‫‪0/881‬‬ ‫‪-‬‬ ‫‪-‬‬ ‫‪-‬‬ ‫‪-‬‬ ‫‪70‬‬ ‫‪-‬‬ ‫‪4-9-1‬‬ ‫‪-‬‬ ‫‪Et-1 , Ra , T,‬‬
‫‪RH‬‬
‫‪0/116‬‬ ‫‪0/827‬‬ ‫‪0/0790‬‬ ‫‪0/890‬‬ ‫‪0/0830‬‬ ‫‪0/883‬‬ ‫‪0/079‬‬ ‫‪0/883‬‬ ‫‪50‬‬ ‫‪50‬‬ ‫‪3-12-1‬‬ ‫‪3-11-1‬‬ ‫‪Et-1 , Ra , T‬‬

‫ﺷﮑﻞ ‪ -۲‬ﻧﺘﺎﻳﺞ ﺭﻭﺵ ‪ MLR‬ﺑﺮﺍﻱ ﺗﺮﮐﻴﺐ ‪ ۶‬ﻣﺘﻐﻴﺮﻩ ﺑﺮﺍﻱ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ‬

‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫‪ ‬‬

 ‫ﺷﮑﻞ ‪ -۳‬ﻧﺘﺎﻳﺞ ﺭﻭﺵ ‪ ANN‬ﺑﺮﺍﻱ ﺗﺮﮐﻴﺐ ‪ ۶‬ﻣﺘﻐﻴﺮﻩ ﺑﺮﺍﻱ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ‬

‫ﺟﺪﻭﻝ ‪ -۳‬ﺧﻼﺻﻪ ﻧﺘﺎﻳﺞ ﮐﻠﻴﻪ ﺭﻭﺷﻬﺎﻱ ﺑﻪﮐﺎﺭ ﺑﺮﺩﻩ ﺷﺪﻩ ﺑﻪﻣﻨﻈﻮﺭ ﺑﺮﺁﻭﺭﺩ ﻣﻘﺪﺍﺭ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﺍﻟﻒ‪ -‬ﺗﺒﺮﻳﺰ ﻭ ﺏ‪ -‬ﺍﺭﻭﻣﻴﻪ‬
‫ﺁﺯﻣﺎﻳﺶ‬ ‫ﺁﻣﻮﺯﺵ‬
‫‪RMSE‬‬ ‫‪2‬‬ ‫‪RMSE‬‬
‫ﺭﻭﺷﻬﺎ‬ ‫)ﺍﻟﻒ(‬
‫)‪(mm/d‬‬ ‫‪R‬‬ ‫)‪(mm/d‬‬ ‫‪R2‬‬
‫‪9/18‬‬ ‫‪0/09‬‬ ‫‪4/06‬‬ ‫‪0/37‬‬ ‫ﺗﻌﺎﺩﻝ ﺍﻧﺮﮊﻱ‬
‫‪4/02‬‬ ‫‪0/38‬‬ ‫‪3/83‬‬ ‫‪0/43‬‬ ‫ﺁﻳﺮﻭﺩﻳﻨﺎﻣﻴﮑﻲ‬
‫‪3/65‬‬ ‫‪0/46‬‬ ‫‪3/20‬‬ ‫‪0/50‬‬ ‫ﺗﺮﮐﻴﺒﻲ ﭘﻨﻤﻦ‬
‫‪0/08‬‬ ‫‪0/89‬‬ ‫‪0/08‬‬ ‫‪0/89‬‬ ‫‪Et-1, RH , Ra , T, P, U2‬‬
‫ﺷﺒﮑﻪ ﻋﺼﺒﻲ‬
‫‪0/09‬‬ ‫‪0/87‬‬ ‫‪0/08‬‬ ‫‪0/87‬‬ ‫‪RH , Ra , T, P, U2‬‬
‫‪1/83‬‬ ‫‪0/86‬‬ ‫‪1/85‬‬ ‫‪0/86‬‬ ‫‪Et-1, RH , Ra , T, P, U2‬‬
‫ﺭﮔﺮﺳﻴﻮﻥ‬
‫‪2/15‬‬ ‫‪0/84‬‬ ‫‪2/15‬‬ ‫‪0/85‬‬ ‫‪RH , Ra , T, P, U2‬‬
‫ﺁﺯﻣﺎﻳﺶ‬ ‫ﺁﻣﻮﺯﺵ‬
‫‪RMSE‬‬ ‫‪2‬‬ ‫‪RMSE‬‬ ‫ﺭﻭﺷﻬﺎ‬ ‫)ﺏ(‬
‫)‪(mm/d‬‬ ‫‪R‬‬ ‫)‪(mm/d‬‬ ‫‪R2‬‬
‫‪9/79‬‬ ‫‪0/004‬‬ ‫‪5/72‬‬ ‫‪0/23‬‬ ‫ﺗﻌﺎﺩﻝ ﺍﻧﺮﮊﻱ‬
‫‪6/01‬‬ ‫‪0/12‬‬ ‫‪3/83‬‬ ‫‪0/41‬‬ ‫ﺁﻳﺮﻭﺩﻳﻨﺎﻣﻴﮑﻲ‬
‫‪4/25‬‬ ‫‪0/35‬‬ ‫‪2/81‬‬ ‫‪0/56‬‬ ‫ﺗﺮﮐﻴﺒﻲ ﭘﻨﻤﻦ‬
‫‪0/11‬‬ ‫‪0/83‬‬ ‫‪0/08‬‬ ‫‪0/89‬‬ ‫‪Et-1, RH , Ra , T, P, U2‬‬
‫ﺷﺒﮑﻪ ﻋﺼﺒﻲ‬
‫‪0/13‬‬ ‫‪0/77‬‬ ‫‪0/09‬‬ ‫‪0/87‬‬ ‫‪RH , Ra , T, P, U2‬‬
‫‪1/61‬‬ ‫‪0/78‬‬ ‫‪1/25‬‬ ‫‪0/85‬‬ ‫‪Et-1, RH , Ra , T, P, U2‬‬ ‫ﺭﮔﺮﺳﻴﻮﻥ‬
‫‪1/974‬‬ ‫‪0/74‬‬ ‫‪1/47‬‬ ‫‪0/82‬‬ ‫‪RH , Ra , T, P, U2‬‬

‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬

‫ﻣﻴﺎﻧﮕﻴﻦ ﺁﻥ ﻭﺭﻭﺩﻱ ﻭ ﺑﺎ ﮐﻤﺘﺮﻳﻦ ﻭﺍﺭﻳﺎﻧﺲ ﻳﺎ ﺍﻧﺤﺮﺍﻑ ﺩﺍﺷﺘﻪ ﺑﺎﺷﻨﺪ‪ .‬ﺑﻪ‬
‫ﺍﻳﻦ ﺗﺮﺗﻴﺐ ﺍﺣﺘﻤـﺎﻝ ﻭﺍﺭﺩ ﺷـﺪﻥ ﺩﺍﺩﻩﻫـﺎﻳﻲ ﮐـﻪ ﻣﻤﮑـﻦ ﺍﺳـﺖ ﺑـﺎ ﺧﻄـﺎ‬
‫ﺍﻧﺪﺍﺯﻩﮔﻴﺮﻱ ﺷﺪﻩ ﺑﺎﺷﻨﺪ ﺩﺭ ﻣﺮﺣﻠﻪ ﺁﻧﺎﻟﻴﺰﺣﺴﺎﺳﻴﺖ ﻣﺪﻝ ﺍﺯ ﺑـﻴﻦ ﺭﻓﺘـﻪ ﻭ‬
‫ﺩﺭ ﻧﺘﻴﺠﻪ ﺗﺤﻘﻴﻖ ﺩﺭ ﺧﺼﻮﺹ ﻣﻴـﺰﺍﻥ ﺣﺴﺎﺳـﻴﺖ ﻣـﺪﻝ ﺑـﻪ ﺁﻥ ﻭﺭﻭﺩﻱ‬
‫ﺧﺎﺹ‪ ،‬ﻧﺘﺎﻳﺞ ﺑﻬﺘﺮﻱ ﺭﺍ ﺗﻮﻟﻴﺪ ﻣﻲﮐﻨﺪ‪ .‬ﺩﺭ ﻣﺮﺣﻠـﻪ ﺑﻌـﺪﻱ ﺑـﺎ ﺍﺳـﺘﻔﺎﺩﻩ ﺍﺯ‬
‫ﻧﻤﻮﻧﻪﻫﺎﻱ ﮔﺮﻓﺘﻪ ﺷﺪﻩ ﻭ ﺭﻭﺍﺑﻂ ﻣﺸﺘﻘﺎﺕ ﺟﺰﺋﻲ ﺍﺭﺍﺋـﻪ ﺷـﺪﻩ‪ ،‬ﺣﺴﺎﺳـﻴﺖ‬
‫ﻧﺴﺒﻲ ﺧﺮﻭﺟﻲ ﺑﻪ ﻭﺭﻭﺩﻱﻫﺎﻱ ﻣﺨﺘﻠﻒ ﺑﻪﺩﺳﺖ ﺁﻣﺪ‪ .‬ﻧﻤـﻮﺩﺍﺭ ﭘـﺮﺍﮐﻨﺶ‬
‫ﻣﻘﺎﺩﻳﺮ ﺣﺴﺎﺳﻴﺖﻫﺎﻱ ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ ﺩﺭ ﺷﻜﻠﻬﺎﻱ ‪ ۴‬ﻭ ‪ ۵‬ﺑﻪﺗﺮﺗﻴﺐ ﺑﺮﺍﻱ‬
‫ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ ﻧﺸﺎﻥ ﺩﺍﺩﻩ ﺷﺪﻩﺍﻧﺪ‪ .‬ﺩﺭ ﺍﻳﻦ ﻧﻤﻮﺩﺍﺭﻫﺎ ﻣﺤﻮﺭ ﺍﻓﻘﻲ ﻧﺸـﺎﻥ‬
‫ﺩﻫﻨﺪﻩ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ﻭ ﻣﺤـﻮﺭ ﻗـﺎﺋﻢ ﻧﺸـﺎﻥ ﺩﻫﻨـﺪﻩ ﻣﻘـﺎﺩﻳﺮ ﺣﺴﺎﺳـﻴﺖ‬
‫ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ )ﻣﺸﺘﻘﺎﺕ ﺧﺮﻭﺟـﻲ ﺗﺒﺨﻴـﺮ ﺭﻭﻱ ﺩﺍﻣﻨـﻪ ﻣﺘﻐﻴـﺮ ﻭﺭﻭﺩﻱ(‬
‫ﻫﺴﺘﻨﺪ‪.‬‬
‫‪ -۲-۵‬ﻧﺘﺎﻳﺞ ﺁﻧﺎﻟﻴﺰﺣﺴﺎﺳﻴﺖ‬
‫ﺑﻪ ﻣﻨﻈﻮﺭ ﺳﻨﺠﺶ ﺣﺴﺎﺳﻴﺖ ‪ ANNs‬ﺁﻣﻮﺯﺵ ﺩﻳـﺪﻩ ﺑـﺎ ﺭﻭﺵ‪ ، PaD‬ﺍﺯ‬
‫ﺳﺎﺧﺘﺎﺭﻫﺎﻱ ﺑﻬﻴﻨﻪ ‪ ANNs‬ﮐﻪ ﺑﺮﺍﻱ ﺗﺮﮐﻴﺒﺎﺕ ‪ ۶‬ﭘﺎﺭﺍﻣﺘﺮﻱ ﺑـﺎ ﺗﻮﺟـﻪ ﺑـﻪ‬
‫ﺟﺪﻭﻝ ‪ ۲‬ﺑﻪ ﺩﺳﺖ ﺁﻣﺪﻧﺪ‪ ،‬ﺍﺳﺘﻔﺎﺩﻩ ﺷﺪ ﻭ ﺩﺭ ﺍﺩﺍﻣﻪ ﻣﺸﺘﻘﺎﺕ ﺗﻮﺍﺑﻊ ﺣـﺎﮐﻢ‬
‫ﺑﺮ ﺷﺒﮑﻪ ﻧﺴﺒﺖ ﺑﻪ ﻫﻤﻪ ﻭﺭﻭﺩﻱﻫﺎ ﻭ ﺑﺎ ﺛﺎﺑﺖ ﻧﮕﻪ ﺩﺍﺷﺘﻦ ﺳـﺎﻳﺮ ﻣﺘﻐﻴﺮﻫـﺎ‬
‫ﺑﻪﺩﺳﺖ ﺁﻣﺪ )ﺗﻐﻴﻴﺮ ﻧﺴﺒﺖ ﺑﻪ ﻭﺭﻭﺩﻱ ﺧﺎﺹ ﺑﺎ ﺛﺎﺑﺖ ﻧﮕﻪ ﺩﺍﺷـﺘﻦ ﺑﻘﻴـﻪ‬
‫ﻭﺭﻭﺩﻱﻫﺎ(‪ .‬ﺳﭙﺲ ﺑﺎ ﺍﺳـﺘﻔﺎﺩﻩ ﺍﺯ ﺑﺎﻧـﮏ ﺍﻃﻼﻋـﺎﺗﻲ ﺩﺍﺩﻩﻫـﺎﻱ ﻣـﺪﻝﻫـﺎ‬
‫)‪ ۵۴۷۹‬ﻭ ‪ ۲۹۲۲‬ﺭﮐﻮﺭﺩ ﺛﺒـﺖ ﺷـﺪﻩ ﺑـﺮﺍﻱ ﻫـﺮ ﻣﺘﻐﻴـﺮ ﻭﺭﻭﺩﻱ( ﻭ ﺩﺭ‬
‫ﻓﻀﺎﻱ‪ ۶‬ﺑﻌـﺪﻱ ﻭﺭﻭﺩﻱﻫـﺎ ﺑـﺮﺍﻱ ﺷـﻬﺮ ﺗﺒﺮﻳـﺰ ﻭ ﺍﺭﻭﻣﻴـﻪ‪ ،‬ﻧﻤﻮﻧـﻪﻫـﺎﻳﻲ‬
‫ﺑﻪﺗﺮﺗﻴﺐ ﺑﺎ ﺗﻌﺪﺍﺩ ‪ ۳۰۰۰‬ﻭ‪ ۱۰۰۰‬ﻧﻘﻄـﻪ ﻭ ﺑـﺎ ﺗﻮﺯﻳـﻊ ﻧﺮﻣـﺎﻝ ﺍﻧﺘﺨـﺎﺏ‬
‫ﺷﺪﻧﺪ‪ .‬ﻧﻤﻮﻧﻪﮔﻴﺮﻱ ﺑﺎ ﺗﻮﺯﻳﻊ ﻧﺮﻣﺎﻝ ﺑﺎﻋﺚ ﻣﻲﺷﻮﺩ ﮐﻪ ﺍﺯ ﺩﺍﻣﻨﻪ ﻫﺮ ﻣﺘﻐﻴﺮ‬
‫ﻭﺭﻭﺩﻱ ﺩﺍﺩﻩﻫﺎﻳﻲ ﺍﻧﺘﺨﺎﺏ ﺷﻮﻧﺪ ﮐﻪ ﺑﻴﺸﺘﺮﻳﻦ ﭘﺮﺍﮐﻨـﺪﮔﻲ ﺭﺍ ﺩﺭ ﺍﻃـﺮﺍﻑ‬

‫ﺷﮑﻞ ‪ -۴‬ﻧﻤﻮﺩﺍﺭﻫﺎﻱ ﭘﺮﺍﮐﻨﺶ ﻣﻘﺎﺩﻳﺮ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﺑﺮﺍﻱ ﺷﻬﺮ ﺗﺒﺮﻳﺰ‬

‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫‪ ‬‬

 ‫ﺷﮑﻞ ‪ -۵‬ﻧﻤﻮﺩﺍﺭﻫﺎﻱ ﭘﺮﺍﮐﻨﺶ ﻣﻘﺎﺩﻳﺮ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﺑﺮﺍﻱ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ‬

‫ﻫﻨﮕﺎﻣﻲ ﻛﻪ ﺍﻳﻦ ﭘﺮﺍﻛﻨﺪﮔﻲ ﻭ ﻳﺎ ﺗﺠﻤﻊ‪ ،‬ﺣﻮﻝ ﺧﻂ ﭘﺎﻳﻪ ﺻﻔﺮ )ﺣﺴﺎﺳﻴﺖ‬
‫ﺻﻔﺮ( ﺑﺎﺷﺪ‪ ،‬ﺍﻫﻤﻴﺖ ﺑﻴﺸﺘﺮﻱ ﭘﻴﺪﺍ ﻣﻲﻛﻨﺪ‪ ،‬ﺑـﻪ ﺍﻳـﻦ ﺗﺮﺗﻴـﺐ ﻛـﻪ ﺗﺠﻤـﻊ‬
‫ﺣــﻮﻝ ﺧــﻂ ﭘﺎﻳـــﻪ ﺻــﻔﺮ‪ ،‬ﻧﺸـــﺎﻧﮕﺮ ﺻــﻔﺮ ﺷـــﺪﻥ ﺗﻘﺮﻳﺒــﻲ ﻣﺸـــﺘﻖ‬
‫‪  f (X1 , X 2 ,...X n ‬ﺩﺭ ﺗﻤﺎﻡ ﻓﻀﺎﻱ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﻭ ﺑﻲﺗـﺄﺛﻴﺮ‬
‫‪‬‬
‫‪‬‬ ‫‪X1‬‬
‫‪‬‬
‫ﺑﻮﺩﻥ ﺗﻐﻴﻴﺮﺍﺕ ﻣﺘﻐﻴﺮ ‪ X1‬ﺑﺮ ﺗﺎﺑﻊ ‪ f‬ﺍﺳﺖ‪ .‬ﺩﺭ ﻋﻴﻦ ﺣﺎﻝ‪ ،‬ﭘﺮﺍﻛﻨﺪﮔﻲ ﺍﻳﻦ‬
‫ﻣﻘﺎﺩﻳﺮ ﺣﻮﻝ ﺧﻂ ﭘﺎﻳﻪ ﺻﻔﺮ ﻧﺸﺎﻥ ﻣـﻲﺩﻫـﺪ ﻛـﻪ ﻣﺘﻐﻴـﺮ ‪ X1‬ﺩﺭ ﺑﻌﻀـﻲ‬
‫ﺳﻄﻮﺡ ﺧﻮﺩ ﻭ ﻳﺎ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺩﻳﮕﺮ ﺗـﺄﺛﻴﺮ ﺍﻓﺰﺍﻳﺸـﻲ ﺑـﺮ ﺗـﺎﺑﻊ ﺩﺍﺭﺩ ﻭ ﺩﺭ‬
‫ﺑﻌﻀﻲ ﺳﻄﻮﺡ ﺩﻳﮕﺮ ﺧﻮﺩ ﻳﺎ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺩﻳﮕـﺮ ﺗـﺄﺛﻴﺮ ﻛﺎﻫﺸـﻲ ﺑـﺮ ﺭﻭﻱ‬
‫ﺗــﺎﺑﻊ ‪ f‬ﺩﺍﺭﺩ‪ ،‬ﺍﺯ ﺍﻳــﻦﺭﻭ ﭘﻴــﺪﺍ ﻛــﺮﺩﻥ ﺍﻧــﺪﺭﻛﻨﺶ ﺑــﻴﻦ ﻣﺘﻐﻴــﺮ ‪ X1‬ﺑــﺎ‬
‫ﻣﺘﻐﻴﺮﻫﺎﻱ ﺩﻳﮕﺮ ﺩﺭ ﺗﺄﺛﻴﺮ ﺑﺮ ﺗﺎﺑﻊ ‪ f‬ﻭ ﺩﺭ ﻧﻈﺮ ﮔﺮﻓﺘﻦ ﺍﻳﻦ ﺍﻧـﺪﺭﻛﻨﺶﻫـﺎ‬
‫ﺩﺭ ﻣﻘﺎﺩﻳﺮ ﻃﺮﺍﺣﻲ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ‪ ،‬ﻣﻲﺗﻮﺍﻧﺪ ﻣﻔﻴﺪ ﺑﺎﺷﺪ‪ .‬ﺑﺮﺍﻱ ﻣﺜﺎﻝ‬
‫ﺩﺭ ﺍﻳﻦ ﻣﻮﺭﺩ ﻣﻲﺗﻮﺍﻥ ﺗﺄﺛﻴﺮ ﺗﺸﻌﺸـﻊ ﻭ ﺗﺒﺨﻴـﺮ ﺭﻭﺯ ﻗﺒـﻞ ﺭﺍ ﺑـﺮ ﺗﺒﺨﻴـﺮ‬
‫ﺭﻭﺯﺍﻧﻪ ﺩﺭ ﺷﻜﻠﻬﺎﻱ ‪ -۶‬ﺍﻟﻒ ﻭ ‪ -۷‬ﺍﻟﻒ ﺑﻪ ﻭﺿﻮﺡ ﻣﻼﺣﻈﻪ ﮐﺮﺩ‪.‬‬
‫ﺑﺮﺍﻱ ﻳﺎﻓﺘﻦ ﺟﻬﺖ ﺗـﺄﺛﻴﺮ ﻭﺭﻭﺩﻱﻫـﺎ ﺑـﺮ ﺧﺮﻭﺟـﻲ ﻣـﺪﻝ ﺷـﺒﮑﻪ‪ ،‬ﺁﻧـﺎﻟﻴﺰ‬
‫ﺩﺭﺻﺪﻫﺎﻱ ﺁﻣﺎﺭﻱ ﻣﻘﺎﺩﻳﺮ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﺑﻪﺩﺳﺖ ﺁﻣـﺪ‪ .‬ﺩﺭ ﺷـﻜﻠﻬﺎﻱ‬
‫‪ ۶‬ﻭ ‪ ۷‬ﻧﻤﻮﺩﺍﺭﻫﺎﻱ ﻧﺘﺎﻳﺞ ﺁﻧﺎﻟﻴﺰ ﺩﺭﺻﺪﻫﺎﻱ ﺁﻣـﺎﺭﻱ ﻣﻘـﺎﺩﻳﺮ ﺣﺴﺎﺳـﻴﺖ‬
‫ﻧﺴــﺒﻲ ﺑــﻪ ﺭﻭﺵ‪ ، PaD‬ﺍﻫﻤﻴــﺖ ﻧﺴــﺒﻲ ﻭﺭﻭﺩﻱﻫــﺎ ﺑــﺎ ﭘــﺎﺭﺍﻣﺘﺮ ‪SSD‬‬
‫‪‬‬
‫ﺑﻪﺭﻭﺵ ‪ ، PaD‬ﺍﻫﻤﻴﺖ ﻧﺴـﺒﻲ ﻭﺭﻭﺩﻱﻫـﺎ ﺑـﻪﺭﻭﺵ ﻭﺯﻥﻫـﺎﻱ ﺍﺗﺼـﺎﻟﻲ‬
‫ﺑﻪﺗﺮﺗﻴﺐ ﺑﺮﺍﻱ ﺷﻬﺮﻫﺎﻱ ﺗﺒﺮﻳﺰ ﻭ ﺍﺭﻭﻣﻴﻪ ﻧﺸﺎﻥ ﺩﺍﺩﻩ ﺷﺪﻩﺍﻧﺪ‪ .‬ﺑﺎ ﺗﻮﺟﻪ ﺑـﻪ‬
‫ﺷـﻜﻠﻬﺎﻱ‪ -۶‬ﺍﻟــﻒ ﻭ‪ -۷‬ﺍﻟــﻒ‪ ،‬ﻻﺯﻡ ﺑـﻪﺫﮐــﺮ ﺍﺳــﺖ ﮐـﻪ ﺗﺠﻤــﻊ ﻣﻘــﺎﺩﻳﺮ‬
‫ﺩﺭﺻﺪﻫﺎﻱ ﺁﻣﺎﺭﻱ ﺣﺴﺎﺳﻴﺖ ﻧﺴﺒﻲ ﻳﮏ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ﻣﺎﻧﻨﺪ ‪ X1‬ﺣﻮﻝ‬
‫ﻳـﮏ ﻋـﺪﺩ ﺛﺎﺑــﺖ‪ ،‬ﺑﻴـﺎﻧﮕﺮ ﺗﻐﻴﻴــﺮﺍﺕ ﺛﺎﺑـﺖ ﺗـﺎﺑﻊ ﻣــﻮﺭﺩ ﻧﻈـﺮ ‪(f(X1,‬‬
‫))‪ X2,…Xn‬ﻧﺴﺒﺖ ﺑﻪ ﻣﺘﻐﻴﺮ ‪ X1‬ﺩﺭ ﻓﻀﺎﻱ ﻣﻮﺭﺩ ﺑﺮﺭﺳﻲ ﺍﺳـﺖ‪ ،‬ﻳﻌﻨـﻲ‬
‫ﺑﺎ ﻗﻄﻌﻴﺖ ﺑﺎﻻ )ﺑﻴﺶ ﺍﺯ ‪ ۹۰‬ﺩﺭﺻﺪ( ﻣﻲﺗﻮﺍﻥ ﮔﻔﺖ ﮐﻪ ﻧﻘﺶ ﻣﺘﻐﻴـﺮ ‪X1‬‬
‫ﺩﺭ ﺗﻐﻴﻴﺮﺍﺕ ﺗﺎﺑﻊ ‪ ، f‬ﻣﺴﺘﻘﻞ ﺍﺯ ﺳﻄﻮﺡ ﺧﻮﺩ ﻣﺘﻐﻴﺮ ‪ X1‬ﻭ ﻧﻴﺰ ﻣﺴـﺘﻘﻞ ﺍﺯ‬
‫ﺗﻐﻴﻴﺮﺍﺕ ﺩﻳﮕﺮ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺍﺳﺖ‪ .‬ﺑﻪﻋﺒﺎﺭﺕ ﺩﻳﮕﺮ ﻣﺘﻐﻴـﺮ ‪ X1‬ﺩﺭ‬
‫ﺗﻐﻴﻴﺮﺍﺕ ﺗﺎﺑﻊ ‪ f‬ﺑﺎ ﻣﺘﻐﻴﺮﻫﺎﻱ ﺩﻳﮕﺮ ﻓﻀـﺎﻱ ﻭﺭﻭﺩﻱ ﺍﻧﺪﺭﮐﻨﺸـﻲ ﻧـﺪﺍﺭﺩ‪.‬‬

‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬

‫ﺷﮑﻞ ‪ -۶‬ﻧﺘﺎﻳﺞ ﺑﺮﺍﻱ ﺷﻬﺮ ﺗﺒﺮﻳﺰ‪ :‬ﺍﻟﻒ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺗﻮﺭﻧﺎﺩﻭ )ﭘﺮﺍﮐﻨﺪﮔﻲ ﺩﺭﺻﺪ ﺣﺴﺎﺳﻴﺖﻫﺎﻱ ﻧﺴﺒﻲ( ﺑﻪﺭﻭﺵ ‪ ، PaD‬ﺏ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺩﺭﺻﺪ ﺍﻫﻤﻴﺖ ﻭ ﺷﺮﮐﺖ‬
‫ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺑﻪﺭﻭﺵ ‪ PaD‬ﻭ ﺑﺎ ﺗﻌﺮﻳﻒ ﭘﺎﺭﺍﻣﺘﺮ ‪ SSD‬ﻭ ﺝ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺩﺭﺻﺪ ﺍﻫﻤﻴﺖ ﻭ ﺷﺮﮐﺖ ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺑﻪﺭﻭﺵ ﻭﺯﻥﻫﺎﻱ ﺍﺗﺼﺎﻟﻲ ﺍﺳﺖ‪.‬‬

‫ﺷﮑﻞ ‪ -۷‬ﻧﺘﺎﻳﺞ ﺑﺮﺍﻱ ﺷﻬﺮ ﺍﺭﻭﻣﻴﻪ‪ :‬ﺍﻟﻒ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺗﻮﺭﻧﺎﺩﻭ )ﭘﺮﺍﮐﻨﺪﮔﻲ ﺩﺭﺻﺪ ﺣﺴﺎﺳﻴﺖﻫﺎﻱ ﻧﺴﺒﻲ ( ﺑﻪﺭﻭﺵ ‪ ، PaD‬ﺏ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺩﺭﺻﺪ ﺍﻫﻤﻴﺖ ﻭ ﺷﺮﮐﺖ‬
‫ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺑﻪﺭﻭﺵ ‪ PaD‬ﻭ ﺑﺎ ﺗﻌﺮﻳﻒ ﭘﺎﺭﺍﻣﺘﺮ ‪ SSD‬ﻭ ﺝ‪ -‬ﻧﻤﻮﺩﺍﺭ ﺩﺭﺻﺪ ﺍﻫﻤﻴﺖ ﻭ ﺷﺮﮐﺖ ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺑﻪﺭﻭﺵ ﻭﺯﻥﻫﺎﻱ ﺍﺗﺼﺎﻟﻲ ﺍﺳﺖ‪.‬‬

‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫‪ ‬‬

‫ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﮐﻪ ﻣﻨﺠﺮ ﺑﻪ ﺗﻌﻴﻴﻦ ﻣﺆﺛﺮﺗﺮﻳﻦ ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺷﺪ‪ ،‬ﻣـﻲﺗـﻮﺍﻥ‬ ‫ﺍﺯ ﺷﮑﻠﻬﺎﻱ ‪ -۶‬ﺍﻟﻒ ﻭ ‪ -۷‬ﺍﻟﻒ ﻣﺸﺨﺺ ﺍﺳﺖ ﻛﻪ ﺣﺴﺎﺳـﻴﺖﻫـﺎﻱ‬
‫ﺑﻪ ﺟﺪﻭﻝ ‪ ۲‬ﻣﺮﺍﺟﻌﻪ ﮐﺮﺩ‪ .‬ﻫﻤﺎﻧﻄﻮﺭﻱ ﮐﻪ ﻣﺸﺎﻫﺪﻩ ﻣﻲﺷﻮﺩ ﭘﺎﺳﺦ ﺷـﺒﮑﻪ‬ ‫ﺧﺮﻭﺟﻲ ﺷﺒﮑﻪ ﺑﺮﺍﻱ ‪ T‬ﺑﻴﺶ ﺍﺯ‪ ۹۰‬ﺩﺭﺻﺪ ﻭ ﺑﺮﺍﻱ ‪ U2‬ﺣﺪﻭﺩ‪ ۹۰‬ﺩﺭﺻﺪ‬
‫ﺯﻣﺎﻧﻲ ﮐﻪ ﺗﻤﺎﻡ ﻣﺘﻐﻴﺮﻫﺎﻱ ﻭﺭﻭﺩﻱ ﺭﺍ ﺩﺭﻳﺎﻓﺖ ﻣﻲﮐﻨـﺪ ﺑـﺎ ﭘﺎﺳـﺦ ﺷـﺒﮑﻪ‬ ‫ﻣﻘﺎﺩﻳﺮ ﻣﺜﺒﺖ ﺍﺳﺖ ﺯﻳﺮﺍ ‪ ۱۰ D‬ﺩﺭﺻﺪ ﺑﺰﺭﮒﺗﺮ ﺍﺯ ﺻﻔﺮ ﺑﻪﺩﺳﺖ ﺁﻣـﺪﻩ‪،‬‬
‫ﺯﻣﺎﻧﻲ ﮐـﻪ ﻓﻘـﻂ ﻭﺭﻭﺩﻱﻫـﺎﻱ ‪ Ra ، T‬ﻭ ‪ Et-1‬ﺭﺍ ﮐـﻪ ﺩﺭ ﻣﺮﺣﻠـﻪ ﺗﺤﻠﻴـﻞ‬ ‫ﺑﺮﺍﻱ ‪ P‬ﻭ ‪ RH‬ﺣﺪﻭﺩ ‪ ۷۵‬ﺩﺭﺻﺪ ﻣﻘﺎﺩﻳﺮ ﻣﻨﻔﻲ ﺑـﻪﺩﺳـﺖ ﺁﻣـﺪﻩ ﻭ ﺩﺭﺻـﺪ‬
‫ﺣﺴﺎﺳﻴﺖ ﻣﺆﺛﺮﺗﺮﻳﻦ ﺗﺸﺨﻴﺺ ﺩﺍﺩﻩ ﺷﺪﻧﺪ‪ ،‬ﺩﺭﻳﺎﻓـﺖ ﻣـﻲﮐﻨـﺪ ﺗﻔـﺎﻭﺕ‬ ‫ﺑﺎﻗﻴﻤﺎﻧﺪﻩ ﺩﺍﺭﺍﻱ ﻣﻘـﺎﺩﻳﺮ ﻣﺜﺒـﺖ ﮐﻮﭼـﮏ ﻫﺴـﺘﻨﺪ‪ .‬ﺑـﺮﺍﻱ ‪ Ra‬ﺣـﺪﻭﺩ‪۵۰‬‬
‫ﭼﻨﺪﺍﻧﻲ ﻧﺪﺍﺭﺩ‪ ،‬ﺑﻨﺎﺑﺮﺍﻳﻦ ﻣﻲﺗـﻮﺍﻥ ﻧﺘﻴﺠـﻪ ﮔﺮﻓـﺖ ﮐـﻪ ﺳـﺎﻳﺮ ﻣﺘﻐﻴﺮﻫـﺎﻱ‬ ‫ﺩﺭﺻﺪ ﻭ ﺑﺮﺍﻱ ‪) Et-1‬ﺗﺒﺨﻴﺮ ﺭﻭﺯ ﻗﺒﻞ( ﺣﺪﻭﺩ ‪ ۷۵‬ﺩﺭﺻﺪ ﻣﻘـﺎﺩﻳﺮ ﻣﺜﺒـﺖ‬
‫ﻭﺭﻭﺩﻱ ﻧﺴــﺒﺖ ﺑــﻪ ﻣﺘﻐﻴﺮﻫــﺎﻱ ‪ Ra ، T‬ﻭ ‪ Et-1‬ﮐــﻢ ﺍﻫﻤﻴ ـﺖ ﺑــﻮﺩﻩ ﻭ ﺩﺭ‬ ‫ﺑــﻪﺩﺳــﺖ ﺁﻣــﺪ‪ ،‬ﺯﻳــﺮﺍ ﺑــﻪﺗﺮﺗﻴــﺐ ‪ ۵۰ D‬ﺩﺭﺻــﺪ ﺑــﺰﺭﮒﺗــﺮ ﺍﺯ ﺻــﻔﺮ ﻭ‬
‫ﺁﻣﻮﺯﺵ ﺷﺒﮑﻪ ﻧﻘﺶ ﮐﻤﻲ ﺍﻳﻔﺎ ﻣﻲﮐﻨﻨﺪ‪ .‬ﻧﺘﻴﺠـﻪﺍﻱ ﮐـﻪ ﮔﺮﻓﺘـﻪ ﻣـﻲﺷـﻮﺩ‬ ‫‪ ۲۵ D‬ﺩﺭﺻﺪ ﺑﺰﺭﮒﺗﺮ ﺍﺯ ﺻﻔﺮ ﻫﺴﺘﻨﺪ‪ .‬ﺍﺯ ﺍﻳﻦﺭﻭ ﺑﻪﻃﻮﺭ ﮐـﻞ ﻣـﻲﺗـﻮﺍﻥ‬
‫ﺣﺎﮐﻲ ﺍﺯ ﺍﻳـﻦ ﺍﺳـﺖ ﮐـﻪ ﺑـﺎ ﺍﻧﺠـﺎﻡ ﺗﺤﻠﻴـﻞ ﺣﺴﺎﺳـﻴﺖ ﻣـﺪﻝ ﺷـﺒﮑﻪ ﺩﺭ‬ ‫ﮔﻔﺖ ﮐﻪ ﺑﺎ ﺍﺣﺘﻤﺎﻝ ‪ ۹۰‬ﺩﺭﺻﺪ ﺑﺎ ﺍﻓﺰﺍﻳﺶ ‪ T‬ﻭ ‪ U2‬ﻣﻘﺪﺍﺭ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧـﻪ‬
‫ﺍﻗﻠﻴﻢﻫﺎﻱ ﻣﺨﺘﻠﻒ ﻣﻲﺗﻮﺍﻥ ﻓﻀـﺎﻱ ﻣﺘﻐﻴﺮﻫـﺎﻱ ﻭﺭﻭﺩﻱ ﺑـﻪ ﺷـﺒﮑﻪ ﻭ ﺩﺭ‬ ‫ﺍﻓﺰﺍﻳﺶ‪ ،‬ﺑﺎ ﺍﺣﺘﻤﺎﻝ ﺣﺪﻭﺩ ‪ ۷۵‬ﺩﺭﺻﺪ ﺑﺎ ﺍﻓﺰﺍﻳﺶ‪ P‬ﻭ ‪ RH‬ﻣﻘـﺪﺍﺭ ﺗﺒﺨﻴـﺮ‬
‫ﻧﺘﻴﺠﻪ ﺯﻣﺎﻥ ﺁﻣﻮﺯﺵ ﺭﺍ ﮐﺎﻫﺶ ﺩﺍﺩ ﻭ ﭘﺎﺳﺨﻲ ﺑﺎ ﺩﻗـﺖ ﻗﺎﺑـﻞ ﻗﺒـﻮﻝ ﻧﻴـﺰ‬ ‫ﺭﻭﺯﺍﻧﻪ ﮐﺎﻫﺶ‪ ،‬ﺑﺎ ﺍﺣﺘﻤﺎﻝ ﺣﺪﻭﺩ ‪ ۵۰‬ﺩﺭﺻﺪ ﺑﺎ ﺍﻓﺰﺍﻳﺶ ‪ Ra‬ﻭ ﺑﺎ ﺍﺣﺘﻤـﺎﻝ‬
‫ﺑﻪﺩﺳﺖ ﺁﻭﺭﺩ‪.‬‬ ‫ﺑـﻴﺶ ﺍﺯ ‪ ۷۵‬ﺩﺭﺻــﺪ ﺑـﺎ ﺍﻓــﺰﺍﻳﺶ ‪ Et-1‬ﻣﻘــﺪﺍﺭ ﺗﺒﺨﻴـﺮ ﺭﻭﺯﺍﻧــﻪ ﺍﻓــﺰﺍﻳﺶ‬
‫ﻣﻲ ﻳﺎﺑﺪ‪ .‬ﺍﺯ ﭘﺮﺍﮐﻨﺪﮔﻲ ﮔﺴﺘﺮﺩﻩ ﺗﺮ ﻣﻘﺎﺩﻳﺮ ﺁﻣﺎﺭﻱ ﺣﺴﺎﺳﻴﺖﻫﺎﻱ ‪Et-1 ،T‬‬
‫‪ -۶‬ﻧﺘﻴﺠﻪﮔﻴﺮﻱ‬ ‫‪ ( ‬ﺣﻮﻝ ﺧﻂ ﺣﺴﺎﺳﻴﺖ ﺻـﻔﺮ ﺩﺭ‬ ‫‪E‬‬ ‫‪‬‬
‫‪‬‬ ‫‪ ‬ﻭ‬ ‫ﻭ ‪E  ،  E  ) Ra‬‬
‫‪ R‬‬ ‫‪ E   T ‬‬
‫ﺷــﺒﮑﻪﻫــﺎﻱ ﻋﺼــﺒﻲ ﻣﺼــﻨﻮﻋﻲ ﺍﺑﺰﺍﺭﻫــﺎﻱ ﻗــﻮﻱ ﺑــﺮﺍﻱ ﻣﻄﺎﻟﻌــﺎﺕ‬ ‫‪ a‬‬ ‫‪‬‬ ‫‪ t 1 ‬‬
‫ﺷﮑﻠﻬﺎﻱ ‪ -۶‬ﺍﻟﻒ ﻭ ‪ -۷‬ﺍﻟﻒ‪ ،‬ﻣﻲﺗﻮﺍﻥ ﻧﺘﻴﺠﻪ ﮔﺮﻓـﺖ ﮐـﻪ ﺍﻳـﻦ ﻣﺘﻐﻴﺮﻫـﺎ‬
‫ﻫﻴﺪﺭﻭﻣﺘﺌﻮﺭﻭﻟﻮﮊﻳﮑﻲ ﻫﺴﺘﻨﺪ ﮐـﻪ ﺩﺭ ﺍﻳـﻦ ﻣﻄﺎﻟﻌـﻪ ﺑﺮﺗـﺮﻱ ﺧـﻮﺩ ﺭﺍ ﺩﺭ‬
‫ﺗﺄﺛﻴﺮ ﺑﻴﺸﺘﺮﻱ ﺭﻭﻱ ﻣﻘﺪﺍﺭ ﺗﺒﺨﻴﺮ ﻧﺴﺒﺖ ﺑﻪ ﺳـﺎﻳﺮ ﻭﺭﻭﺩﻱﻫـﺎ ﺩﺍﺭﻧـﺪ‪ .‬ﺍﺯ‬
‫ﻣﺴﺌﻠﻪ ﭘﻴﺶﺑﻴﻨﻲ ﺗﺒﺨﻴﺮ ﻧﺴﺒﺖ ﺑﻪ ﺳﺎﻳﺮ ﺭﻭﺷﻬﺎﻱ ﺑـﻪﮐـﺎﺭ ﺑـﺮﺩﻩ ﺷـﺪﻩ ﺑـﻪ‬
‫ﺷــﮑﻠﻬﺎﻱ ‪ -۶‬ﺏ ﻭ ‪ -۶‬ﺝ ﻭ ‪ -۷‬ﺏ ﻭ ‪ -۷‬ﺝ ﻧﻴــﺰ ﻣﻌﻠــﻮﻡ ﺍﺳــﺖ ﮐــﻪ‬
‫ﺍﺛﺒﺎﺕ ﺭﺳﺎﻧﺪﻧﺪ‪ .‬ﻋﻴﺐ ﺍﻳﻦ ﺷـﺒﮑﻪﻫـﺎ ﺩﺭ ﻏﻴﺮﻗﺎﺑـﻞ ﺗﻔﺴـﻴﺮ ﺑـﻮﺩﻥ ﻧﺘـﺎﻳﺞ‬
‫ﺑﻴﺸﺘﺮﻳﻦ ﺗﺄﺛﻴﺮ ﺭﺍ ﭘﺎﺭﺍﻣﺘﺮ ‪ T‬ﻭ ﭘﺲ ﺍﺯ ﺁﻥ ﺑـﻪﺗﺮﺗﻴـﺐ ‪ Et-1‬ﻭ ‪ Ra‬ﺩﺍﺭﻧـﺪ‪.‬‬
‫ﺣﺎﺻﻞ ﺍﺯ ﻣﺪﻝﺳﺎﺯﻱ ﺍﺯ ﻧﻈﺮ ﻣﻴﺰﺍﻥ ﺗﺄﺛﻴﺮ ﻫﺮ ﻣﺘﻐﻴﺮ ﻭﺭﻭﺩﻱ ﺑﺮ ﺧﺮﻭﺟﻲ‬
‫ﺍﺯ ﺭﻭﻱ ﺷــﮑﻠﻬﺎﻱ ‪ ۴‬ﻭ ‪ ۵‬ﻣــﻲﺗــﻮﺍﻥ ﭘﺮﺍﮐﻨــﺪﮔﻲ ﻣﻘــﺎﺩﻳﺮ ﺣﺴﺎﺳــﻴﺖ ﺭﺍ‬
‫ﻣﻮﺭﺩ ﻧﻈﺮ ﺍﺳﺖ‪ .‬ﺍﺯ ﺍﻳﻦﺭﻭ ﺩﺭ ﺍﻳﻦ ﻣﻄﺎﻟﻌﻪ ﺑﺎ ﺍﻧﺠﺎﻡ ﺁﻧﺎﻟﻴﺰ ﺣﺴﺎﺳﻴﺖ ﺑـﻪ‬
‫ﻣﻼﺣﻈﻪ ﮐﺮﺩ ﮐﻪ ﺑـﺮﺍﻱ ﺑﻴﺸـﺘﺮ ﻣﻘـﺎﺩﻳﺮ ‪ T‬ﻭ ‪ U2‬ﺩﺭ ﺑـﺎﻻﻱ ﺧـﻂ ﺻـﻔﺮ‪،‬‬
‫ﺩﻭ ﺭﻭﺵ ﻣﺸﺘﻘﺎﺕ ﺟﺰﺋﻲ ﻭ ﻭﺯﻥﻫﺎﻱ ﺍﺗﺼﺎﻟﻲ‪ ،‬ﺍﻫﻤﻴـﺖ ﻧﺴـﺒﻲ ﺷـﺮﮐﺖ‬
‫ﺑﺮﺍﻱ ﺑﻴﺸﺘﺮ ﻣﻘـﺎﺩﻳﺮ ‪ P‬ﻭ ‪ RH‬ﺩﺭ ﭘـﺎﻳﻴﻦ ﺧـﻂ ﺻـﻔﺮ‪ ،‬ﺑـﺮﺍﻱ ‪ Et-1‬ﺑﻴﺸـﺘﺮ‬
‫ﻭﺭﻭﺩﻱ ﻫﺎﻱ ﺷﺒﮑﻪ ﺩﺭ ﺑﺮﺁﻭﺭﺩ ﺗﺒﺨﻴﺮ ﺭﻭﺯﺍﻧﻪ ﺷﻬﺮﻫﺎﻱ ﺗﺒﺮﻳـﺰ ﻭ ﺍﺭﻭﻣﻴـﻪ‬
‫ﺗﻮﺯﻳﻊﻫﺎ ﺩﺭ ﺑﺎﻻﻱ ﺧﻂ ﺻﻔﺮ ﻭ ﺑﺮﺍﻱ ‪ ، Ra‬ﺣﺴﺎﺳﻴﺖﻫـﺎ ﺗﻮﺯﻳـﻊ ﺗﻘﺮﻳﺒـﺎﹰ‬
‫ﺑﻪﺩﺳﺖ ﺁﻣﺪ ﮐﻪ ﻫﺮ ﺩﻭ ﺭﻭﺵ ﻧﺘﺎﻳﺞ ﺗﻘﺮﻳﺒﺎﹰ ﻫﻤﺴﻮﻳﻲ ﺭﺍ ﻧﺸﺎﻥ ﻣﻲﺩﻫﻨـﺪ‪.‬‬
‫ﻣﺴﺎﻭﻱ ﺩﺭ ﻃﺮﻓﻴﻦ ﺧﻂ ﺣﺴﺎﺳﻴﺖ ﺻﻔﺮ ﺩﺍﺭﻧﺪ‪.‬‬
‫ﺑﺮﺍﻱ ﻳﺎﻓﺘﻦ ﺟﻬﺖ ﺗﺄﺛﻴﺮ ﻫﺮ ﻭﺭﻭﺩﻱ ﺩﺭ ﺧﺮﻭﺟﻲ ﺷﺒﮑﻪﻫﺎ‪ ،‬ﺭﻭﺵ ﺁﻣﺎﺭﻱ‬
‫ﻃﺒــﻖ ﻧﻤﻮﺩﺍﺭﻫــﺎﻱ ﺗﻮﺭﻧــﺎﺩﻭ )ﭘﺮﺍﮐﻨــﺪﮔﻲ ﺩﺭﺻــﺪ ﺣﺴﺎﺳــﻴﺖﻫــﺎﻱ‬
‫ﺁﻧﺎﻟﻴﺰ ﺩﺭﺻﺪ ﺣﺴﺎﺳﻴﺖﻫـﺎﻱ ﻧﺴـﺒﻲ ﺩﺭ ﺭﻭﺵ ‪ PaD‬ﺑـﻪﮐـﺎﺭ ﺑـﺮﺩﻩ ﺷـﺪ‪.‬‬
‫ﻧﺴﺒﻲ( ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ ‪ ،‬ﻣﻴﺰﺍﻥ ﺗﺄﺛﻴﺮ ﺩﻣﺎ‪ ،‬ﺗﺸﻌﺸﻊ ﻭ ﺗﺒﺨﻴـﺮ ﺭﻭﺯ ﻗﺒـﻞ ﺩﺭ‬
‫ﻃﺒﻖ ﻧﻤﻮﺩﺍﺭﻫﺎﻱ ﺗﻮﺭﻧﺎﺩﻭ )ﭘﺮﺍﮐﻨﺶ ﺩﺭﺻﺪﻫﺎﻱ ﺣﺴﺎﺳﻴﺖﻫﺎﻱ ﻧﺴـﺒﻲ(‬
‫ﻫﺮ ﺩﻭ ﻣﻨﻄﻘﻪ ﺑﻴﺸﺘﺮ ﺍﺯ ﻣﺎﺑﻘﻲ ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺍﺳﺖ‪ ،‬ﻭﻟﻲ ﻣﻴﺰﺍﻥ ﺗﺄﺛﻴﺮ ﺩﻣـﺎ ﺩﺭ‬
‫ﺑﻪﺩﺳﺖ ﺁﻣﺪﻩ‪ ،‬ﻣﻴﺰﺍﻥ ﺗﺄﺛﻴﺮ ﺩﻣﺎ‪ ،‬ﺗﺸﻌﺸﻊ ﻭ ﺗﺒﺨﻴـﺮ ﺭﻭﺯ ﻗﺒـﻞ ﺩﺭ ﻫـﺮ ﺩﻭ‬
‫ﺷﻬﺮ ﺗﺒﺮﻳﺰ ﺑﻴﺸﺘﺮ ﺍﺯ ﺍﺭﻭﻣﻴﻪ ﺍﺳﺖ )ﺷﮑﻠﻬﺎﻱ ‪ -۶‬ﺍﻟـﻒ ﻭ ‪ -۷‬ﺍﻟـﻒ(‪ .‬ﺑﻨـﺎ‬
‫ﻣﻨﻄﻘﻪ ﺑﻴﺸﺘﺮ ﺍﺯ ﻣﺎﺑﻘﻲ ﭘﺎﺭﺍﻣﺘﺮﻫﺎ ﺑﻮﺩﻩ ﻭ ﺗﺄﺛﻴﺮ ﺳﺎﻳﺮ ﻋﻮﺍﻣﻞ ﻧﺎﻣﺤﺴـﻮﺱ‬
‫ﺑــﻪ ﻧﺘــﺎﻳﺞ ﺁﻧــﺎﻟﻴﺰ ﺣﺴﺎﺳــﻴﺖ‪ ،‬ﺗــﺄﺛﻴﺮ ﺳــﺎﻳﺮ ﻋﻮﺍﻣــﻞ ﺑــﺮ ﻣﻴــﺰﺍﻥ ﺗﺒﺨﻴــﺮ‬
‫ﺍﺳﺖ‪.‬‬
‫ﻧﺎﻣﺤﺴﻮﺱ ﺍﺳﺖ‪ .‬ﺑـﺮﺍﻱ ﻣﻘﺎﻳﺴـﻪ ﭘﺎﺳـﺦ ‪ ANNs‬ﺑـﺎ ﻧﺘـﺎﻳﺞ ﺣﺎﺻـﻞ ﺍﺯ‬

‫‪ -۷‬ﻣﻨﺎﺑﻊ‬
‫)‪1- Alizadeh, A. (2005). Principles of applied hydrology, University of Mashhad, Iran. (In Persian‬‬
‫‪2- Zoqi, M.J., and Saeedi, M. (2011). “Modeling leachate generation using artificial neural networks.” J. of‬‬
‫)‪Water and Wastewater, 77, 76-84. (In Persian‬‬
‫‪3- Tabesh, M., and Dini, M. (2010). “Forecasting daily urban water demand using artificial neural networks, a‬‬
‫)‪case study of Tehran urban water.” J. of Water and Wastewater, 73, 84-95. (In Persian‬‬
‫‪4- Lu, M., AbouRizk, S.M., and Hermann, U.H. (2001). “Sensitivity analysis of neural networks in spool‬‬
‫‪fabrication productivity studies.” J. of Comput. Civ. Eng, 15, 299-308.‬‬
‫‪5- Li, Y., Shen, L.Y., and Love, P.E.D. (1999). “ANN based mark-up estimation system with self-explanatory‬‬
‫‪capabilities.” J. of Construct. Eng. Manag., 125, 185-189.‬‬

‫‪‬‬ ‫ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ ‫ﺷﻤﺎﺭﻩ ‪ ٣‬ﺳﺎﻝ ‪١٣٩٢‬‬

6- Sinha, S.K., and Mckin, R.A. (2000). “Artificial neural network for measuring organizational effectiveness.”
J. of Comput. Civ. Eng., 14, 9-14.
7- Knoweles, P. (1997). “Predicting labor productivity using neural networks.” M.Sc. Thesis, University of
Alberta, Edmonton, Canada.
8- Sayed, T., and Razvi, A. (2000). “Comparision of neural and conventional approaches to mode choise
analysis.” J. of Comput. Civ. Eng., 14, 23-30.
9- Jain, S.K., Nayak, P.C., and Sudheer, K.P. (2008). “Models for estimating evapotranspiration using artificial
neural networks, and their physical interpretation.” J. of Hydrol. Process, 22, 2225-2234.
10- Moghaddamnia, A., Ghafari Gousheh, M., Piri, J., Amin, S., and Han, D. (2009). “Evaporation estimation
using artificial neural networks and adaptive neuro-fuzzy inference system techniques.” J. of Adv. Water
Resour., 32, 88-97.
11- Deswal, S., and Pal, M. (2008). “Artificial neural network based modeling of evaporation losses in
reservoirs.” J. of Eng. Tech. Manag., 39, 79-83.
12- Rahimikhoob, A. (2008). “Estimating daily pan evaporation using artificial neural network in a semi-arid
environment.” J. of Theor. Appl. Climatol, 98, 101-105.
13- Traore, S., Wang, Y.M., and Kerh, T. (2010). “Artificial neural network for modeling reference
evapotranspiration complex process in Sudano-Sahelian zone.” J. of Agr. Water Manag., 97, 707-714.
14- Ma, L., Wu, Y., Ji, J., and He, Ch. (2011). “The prediction model for soil water evaporation based on BP
neural network.” Int Conf Intell Comput Tech Autom (ICICTA), 2, 276-280.
15- Tabari, H., Marofi, S., and Savziparvar, A.A. (2010). “Estimation of daily pan evaporation using artificial
neural networks and multivariate non-linear regression.” J. of Irrig. Sci., 28, 399-406. (In Persian)
16- Banimahd, M., Yasrobi, S.S., and Woodward, P.K. (2005). “Artificial neural network for stress-strain
behavior of sandy soils: Knowledge based verification.” J. of Comput. Geotech., 32, 377-386.
17- Jain, A., Kumar, S., and Misra, S. (2008). “Modeling and analysis of concrete slump using artificial neural
networks.” J. of Mater. Civ. Eng., 20, 628-633.
18- Gevrey, M., Dimopoulos, I., and Lek, S. (2006). “Two-way interaction of input variables in the sensitivity
analysis of neural network models.” J. of Ecol. Model., 195, 43-50.
19- Liong, S.Y., Lim, W.H., and Paudyal, G.N. (2000). “River stage forecasting in Bangladesh: Neural network
approach.” J. of Comput. Civ. Eng., 14, 1-8.
20- Chow, V.T., Maidment, D.R., and Mays, L.W. (1988). Applied hydrology, McGraw-Hill Book Company,
New York.
21- Phien, N.H., and Kha, N.D.A. (2003). “Flood forecasting for the upper reach of the Red river basin, North
Vietnam.” J. of Water SA., 29, 267-272.
22- Reddy, T.A. (2006). Evaluation and assessment of fault detection and diagnostic methods for centrifugal
chillers-phase II, Drexel University, Philadelphia.
23- Soltani, A. (1998). Applied regression analysis, Isfahan University Pub., Isfahan, Iran. (In Persian)
24- Rajaee, T., Mirbagheri, S.A., Nourani, V., and Alikhani, A. (2010). “Prediction of daily suspended sediment
load using wavelet and neurofuzzy combined model.” Int J. of Environ. Sci. Tech., 7, 93-110
25- Garson, G.D. (1991). “Interpreting neural network connection weights.” J. of Artif. Intell. Expert, 6, 47-51.
26- Goh, A.T.C. (1995). “Back-propagation neural networks for modeling complex systems.” J. of Artif. Intell.
Eng., 9, 143-151.
27- Gevrey, M., Dimopoulos, I., and Lek, S. (2003). “Review and comparison of methods to study the
contribution of variables in artificial neural network models.” J. of Ecol. Model, 160, 249-264.

١٣٩٢ ‫ ﺳﺎﻝ‬٣ ‫ﺷﻤﺎﺭﻩ‬ ‫ ﺁﺏ ﻭ ﻓﺎﺿﻼﺏ‬ 