TN231 : Reserve Estimation of Delkan Iron Ore Deposit by Using Artificial Neural Network
Thesis > Central Library of Shahrood University > Mining, Petroleum & Geophysics Engineering > MSc > 2013
Authors:
Ali Farhadi [Author], Reza Khalou Kakaie[Supervisor], Farhang Sereshki[Supervisor]
Abstarct: One of the main purposes of exploration operations is reserve evaluation. After this stage the economically extraction can studied. In this study the reserve of Delkan iron ore deposit was estimated by two different methods, mean square distance and artificial neural network method respectively. For reserve estimation of deposit, in the first step the 3D model of deposit was made by data of exploration holes. In next step the block model of deposit was produced by using Datamine Studio software. After primary study on input data, the reserve was estimated by mean square distance method and using this method for cutoff grade 19% and density equal to 4.3 (ton/m3) the value of reserve and mean grade of Fe was obtained 1474 thousand ton and 31.94% respectively. In the second step the reserve was estimated by artificial neural network. For this purpose the back-propagation network and the radial basis network were used. In this study 2 approaches has been considered. In first approach all 14 exploration holes has been divided to two parts randomly: training boreholes and testing boreholes. In this case data of training boreholes has been trained the networks and data of testing boreholes has been used for test the networks. For second approach, for training and testing the networks, 70% and 30% of all data from all boreholes was used respectively. Finally by comparing the correlation coefficient and mean square error (RMS) was obtained from networks; it was shown that the radial basis network has worst result than the back-propagation network in the both states. Also back-propagation network has better result in the second approach. A four-laxyer network was found to be optimum back-propagation algorithm in the second state with architecture of 18-15-1. The activation functions of hidden laxyers were tangent hyperbolic. The correlation coefficients of this network were 0.97 and 0.75 for training and testing dataset. Accordingly, by using this method the value of reserve and mean grade of Fe for cutoff grade 19% in the deposit was obtained 1960 thousand ton and 44.5% respectively. After reserve estimation, the production scheduling was done for this deposit. Accordingly the obtained results, mine life was calculated about 6 years in 3 pushbacks and by 6 extraction phases with 431billion Rial net present value.
Keywords:
#Reserve Estimation #Datamine software #Artificial Neural Network (ANN) #Production Scheduling Link
Keeping place: Central Library of Shahrood University
Visitor: