استفاده از شبكه عصبي مصنوعي به منظور مدلسازي فرايند بيوليچينگ فلزات با ارزش از خاكستر سوخت نفت كوره با استفاده از باكتري اسيدي تيوباسيلوس فرواكسيدانس

Using Artificial Neural Network to modeling of valuable metals bioleaching from fuel oil fly ash using Acidithiobacillus ferrooxidans

در اﯾﻦ ﻣﻄﺎﻟﻌﻪ ﻣﺪلﺳﺎزي ﺑﯿﻮﻟﯿﭽﯿﻨﮓ ﻓﻠﺰات ﺑﺎارزش واﻧﺎدﯾﻮم، ﻧﯿﮑﻞ و ﻣﺲ ﻣﻮﺟﻮد در ﺧﺎﮐﺴﺘﺮﻫﺎي ﺳﻮﺧﺖ ﻧﻔﺖﮐﻮره ﺑﺎ اﺳﺘﻔﺎده ﺷﺒﮑﻪﻫﺎي ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ ﺑﺮرﺳﯽ ﻣﯽﺷﻮد. در ﻣﺪلﻫﺎي ﺑﻪدﺳﺖآﻣﺪه، درﺻﺪ اﺳﺘﺨﺮاج ﻓﻠﺰات ﺑﻪﻋﻨﻮان ﺗﺎﺑﻌﯽ از ﻓﺎﮐﺘﻮرﻫﺎي pH )در ﺑﺎزه 1- 2/5(، ﻏﻠﻈﺖ اوﻟﯿﻪ ﯾﻮن Fe2 )در ﺑﺎزه 0- 9 ﮔﺮم ﺑﺮ ﻟﯿﺘﺮ(، درﺻﺪ ﺗﻠﻘﯿﺢ ﺑﺎﮐﺘﺮي )در ﺑﺎزه 1- 10 %( و زﻣﺎن )در ﺑﺎزه 0- 15 روز( ﻓﺮاﯾﻨﺪ ﺑﺮرﺳﯽ ﺷﺪه اﺳﺖ. ﺳﻪ ﻣﺪل ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﺑﺮاي ﺗﺨﻤﯿﻦ درﺻﺪ اﺳﺘﺨﺮاج ﻫﺮﯾﮏ از ﻓﻠﺰات اراﺋﻪ ﺷﺪ. از روش ﭘﺲ اﻧﺘﺸﺎر ﺧﻄﺎ و اﻟﮕﻮرﯾﺘﻢ ﻟﻮﻧﺒﺮگ-ﻣﺎرﮐﻮرت ﺑﺮاي آﻣﻮزش ﺷﺒﮑﻪ اﺳﺘﻔﺎده ﺷﺪ. ﯾﮏﭼﻬﺎرم دادهﻫﺎ در ﻓﺮاﯾﻨﺪ آﻣﻮزش ﺷﺒﮑﻪ ﻋﺼﺒﯽ اﺳﺘﻔﺎده ﻧﺸﺪ و ﺑﺮاي ارزﯾﺎﺑﯽ ﻣﺪل ﻣﻮرد اﺳﺘﻔﺎده ﻗﺮار ﮔﺮﻓﺖ. ﻣﺘﻮﺳﻂ ﺧﻄﺎي ﻧﺴﺒﯽ )MRE( ﺑﺮاي واﻧﺎدﯾﻮم، ﻧﯿﮑﻞ و ﻣﺲ ﺑﻪﺗﺮﺗﯿﺐ ﺑﺮاﺑﺮ ﺑﺎ % 3/07 % ،5/35 و % 2/82 ﺑﻪدﺳﺖ آﻣﺪ. ﻫﻤﭽﻨﯿﻦ ﻣﻘﺪار ﺑﺰرگﺗﺮ از 0/99 از ﮐﺴﺮ ﻣﻄﻠﻖ وارﯾﺎﻧﺲ )R2( ﺑﯿﺎنﮔﺮ ﺗﺄﺋﯿﺪ اﻋﺘﺒﺎر ﻣﺪلﻫﺎي ﺑﻪدﺳﺖ آﻣﺪه از ﺷﺒﮑﻪ ﻋﺼﺒﯽ اﺳﺖ.

In this study, the modeling of vanadium, nickel and copper bioleaching from fuel oil ash ash using artificial neural networks was investigated. In the obtained models, the extraction percentage of metals was investigated as a function of factors such as initial pH (from 1-2.5), initial Fe2+ concentration (from 0 – 9 g/l), initial bacterial inoculation (from 1 – 10%) and process time (from 0-15 day). Three neural network models were presented to estimate the extraction percentage of metals. The propagation error method and Levenberg–Marquardt algorithm were used for training. Furthermore, trial and error method was used to determine the optimal number of neurons. One quarter of the data were used to evaluate the model and were not used for training process. The Mean Relative Errors (MRE) were obtained 5.35%, 3.07% and 2.82% for V, Ni and Cu, respectively. Also the higher 0.99 of R2 indicates the validity of the obtained models.