Intelligent prediction of heating value of coal

Subject Areas : Mineralogy

A. K.  Verma ¹ (Center for Research on Energy Security, The Energy and Resources Institute, IHC Complex, Lodhi Road, New Delhi - 110 003, India)
T. N.  Singh ² (Department of Earth Science, Indian Institute of Technology, Powai, Bombay-76, India)
M.  Monjezi ³ (Department of Mining Engineering, Tarbiat Modares University, Iran)

Received: 2016-05-28 Accepted : 2016-05-28 Published : 2010-03-21

Keywords: Genetic Algorithm, Sensitivity analysis, Neural Networks, Hidden layer, GCV, CANFIS, Epoch,

Abstract :

The gross calorific value (GCV) or heating value of a sample of fuel is one of the important properties which defines the energy of the fuel. Many researchers have proposed empirical formulas for estimating GCV value of coal. There are some known methods like Bomb Calorimeter for determining the GCV in the laboratory. But these methods are cumbersome, costly and time consuming. In this paper, multivariate regression analysis and Co-active neuro-fuzzy inference system (CANFIS) backed by genetic algorithm technique is used for the prediction of GCV, taking all the major constituents of the proximate and ultimate analyses properties as input parameters and the suitability of one technique over the other has been proposed based on the results. Correlations have been developed using multivariate regression analysis that are simple to use based on the proximate and ultimate analysis of data sets from 25 different states of USA because a very through study has been done and the data available is less variable. Also, CANFIS backed by genetic algorithm model is designed to predict the GCV of 4540 US coal samples from the abovementioned datasets. Optimization of the network architecture is done using a systematic approach (genetic algorithm). The network was trained with 4371, cross validation with 100, predicted with rest 69 datasets and the predicted results were compared with the observed values. The mean average percentage error in prediction is found to be negligible (0.2913%) and the generalization capability of the model was established to be excellent. A useful concept of sensitivity analysis is adopted to set the hierarchy of influence of input factors. The results of the present investigation provide functional and vital information for prediction of GCV of any type of coal in USA.