ﻣﻘﺎﯾﺴﻪ ﮐﺎراﯾﯽ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و رﮔﺮﺳﯿﻮن در ﭘﯿﺶﺑﯿﻨﯽ زﻣﺎن ﭼﻮﺑﮑﺸﯽ اﺳﮑﯿﺪر ﭼﺮخ زﻧﺠﯿﺮي زﺗﻮر و ﺗﺮاﮐﺘﻮر ﮐﺸﺎورزي

Comparison of the efficiency of artificial neural network and regression in predicting the skidding time of steel-tracked skidder and agriculture tractor

داﺷﺘﻦ اﻃﻼﻋﺎت دﻗﯿﻖ درﺑﺎره ﺑﺎزدﻫﯽ ﻣﺎﺷﯿﻦآﻻت ﭼﻮﺑﮑﺸﯽ ﺑﻪﻣﻨﻈﻮر ﮐﺎﻫﺶ ﻫﺰﯾﻨﻪﻫﺎي ﺣﻤﻞوﻧﻘﻞ ﺑﺎ اﺳﺘﻔﺎده از ﻣﺪلﻫﺎي آﻣﺎري ﻧﻮﯾﻦ در ﻣﻄﺎﻟﻌﻪ-ﻫﺎي ﻣﻬﻨﺪﺳﯽ ﺟﻨﮕﻞ ﺑﺴﯿﺎر ﺑﺎارزش اﺳﺖ. در اﯾﻦ ﻣﻄﺎﻟﻌﻪ ﻣﺪلﺳﺎزي ﭘﯿﺶﺑﯿﻨﯽ زﻣﺎن ﭼﻮﺑﮑﺸﯽ اﺳﮑﯿﺪر ﭼﺮخ زﻧﺠﯿﺮي زﺗﻮر و ﺗﺮاﮐﺘﻮر ﮐﺸﺎورزي ﺑﺎ اﺳﺘﻔﺎده از ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و ﻣﺪل رﮔﺮﺳﯿﻮن ﺧﻄﯽ ﭼﻨﺪﮔﺎﻧﻪ اﻧﺠﺎم ﺷﺪ و ﺳﭙﺲ ﮐﺎرآﯾﯽ ﻣﺪلﻫﺎ ﺑﺎ ﻫﻢ ﻣﻘﺎﯾﺴﻪ ﮔﺮدﯾﺪ. ﻣﺘﻐﯿﺮﻫﺎي ﻓﺎﺻﻠﻪ ﭼﻮﺑﮑﺸﯽ، ﺷﯿﺐ ﻣﺴﯿﺮ ﭼﻮﺑﮑﺸﯽ، ﺣﺠﻢ و ﺗﻌﺪاد ﮔﺮدهﺑﯿﻨﻪ در ﻫﺮ ﻧﻮﺑﺖ ﭼﻮﺑﮑﺸﯽ ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮﻫﺎي ﻣﺴﺘﻘﻞ )ﻣﺘﻐﯿﺮ ورودي( و زﻣﺎن ﻫﺮ ﻧﻮﺑﺖ ﭼﻮﺑﮑﺸﯽ ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮ واﺑﺴﺘﻪ )ﻣﺘﻐﯿﺮ ﭘﺎﺳﺦ( وارد ﻣﺪل ﺷﺪﻧﺪ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد ﮐﻪ در ﭘﯿﺶﺑﯿﻨﯽ زﻣﺎن ﭼﻮﺑﮑﺸﯽ اﺳﮑﯿﺪر ﭼﺮخ زﻧﺠﯿﺮي زﺗﻮر ﻣﯿﺰان ﺿﺮﯾﺐ ﺗﺒﯿﯿﻦ ﺷﺒﮑﻪ ﻋﺼﺒﯽ MLP و ﻣﺪل رﮔﺮﺳﯿﻮن ﺑﻪ ﺗﺮﺗﯿﺐ 0/78 و 0/55 و ﻣﯿﺰان ﺧﻄﺎي ﻣﺪلﻫﺎ ﺑﻪ ﺗﺮﺗﯿﺐ 0/19 و 0/42 ﻣﯽﺑﺎﺷﺪ. ﻫﻤﭽﻨﯿﻦ در ﺳﯿﺴﺘﻢ ﭼﻮﺑﮑﺸﯽ ﺗﺮاﮐﺘﻮر ﮐﺸﺎورزي ﻣﯿﺰان ﺿﺮﯾﺐ ﺗﺒﯿﯿﻦ ﺷﺒﮑﻪ ﻋﺼﺒﯽ MLP و ﻣﺪل رﮔﺮﺳﯿﻮن ﺑﻪ ﺗﺮﺗﯿﺐ 0/70 و 0/62 و ﻣﯿﺰان ﺧﻄﺎي ﻣﺪلﻫﺎ ﺑﻪ ﺗﺮﺗﯿﺐ 0/18 و 0/28 ﻣﯽﺑﺎﺷﺪ. ﺑﻨﺎﺑﺮاﯾﻦ در ﻫﺮ دو ﺳﯿﺴﺘﻢ ﭼﻮﺑﮑﺸﯽ ﺷﺒﮑﻪ ﻋﺼﺒﯽ MLP در ﭘﯿﺶﺑﯿﻨﯽ زﻣﺎن ﭼﻮﺑﮑﺸﯽ ﻧﺴﺒﺖ ﺑﻪ ﻣﺪل رﮔﺮﺳﯿﻮن ﺧﻄﯽ ﭼﻨﺪﮔﺎﻧﻪ ﮐﺎرآﯾﯽ ﺑﯿﺸﺘﺮي دارد. ﺗﺤﻠﯿﻞ ﺣﺴﺎﺳﯿﺖ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ و رﮔﺮﺳﯿﻮن ﻧﺸﺎن داد ﮐﻪ در اﺳﮑﯿﺪر ﭼﺮخ زﻧﺠﯿﺮي زﺗﻮر ﻣﺘﻐﯿﺮ ﻓﺎﺻﻠﻪ ﭼﻮﺑﮑﺸﯽ و در ﺗﺮاﮐﺘﻮر ﮐﺸﺎورزي ﻣﺘﻐﯿﺮ ﺷﯿﺐ ﻣﺴﯿﺮ ﭼﻮﺑﮑﺸﯽ ﺑﯿﺸﺘﺮﯾﻦ اﻫﻤﯿﺖ را دارﻧﺪ.

Having accurate information about the efficiency of skidding machines in order to reduce transportation costs in forest engineering studies using modern statistical models is very valuable. In this study, the prediction of the skidding time in steel tracked skidder and agriculture tractor was performed using an artificial neural network and multiple linear regression model and then the efficiency of the models was compared. The variables of skidding distance, slope, and volume in each skidding cycle as independent variables (input variable) and time of each skidding cycle as the dependent variables (response variable) were entered into the model. The results showed the prediction in skidding time of steel tracked skidder, the explanation coefficient of the MLP neural network and regression model were 0.78 and 0.55, respectively and the error rate of models was 0.19 and 0.42, respectively. Also, in the agricultural tractor system, the explanation coefficient of MLP neural network and regression model were 0.70 and 0.62, respectively, and the error rate of models was 0.18 and 0.28, respectively. Therefore, in both skidding systems, MLP neural network is more efficient in predicting skidding time than the multiple linear regression model. Sensitivity analysis of the artificial neural network and regression showed that the skidding distance variable in the steel tracked skidder chain wheel and the skidding path slope variable in the agricultural tractor are the most important.