Estimation of the coal higher heating value for energy systems relied on ultimate analysis with machine learning techniques

The calorific value of solid fuels, also referred to as the gross calorific value (GCV) or the higher heating value (HHV), is a crucial property for its use as a fuel in energy systems. The HHV of coal as a resource can be predicted by more effective algorithms that use schedule information in engineering, like ultimate analysis, enabling fast decisions about its use as fuel in energy systems. The goal of this research was to acquire a global artificial prediction model relied on an interesting algorithm, a nonlinear model termed multivariate adaptive regression splines (MARS), in addition to the grid search (GS) optimizer, for characterization of coal HHV (output variable) using constituents of coal ultimate analysis: carbon (C), nitrogen (N), oxygen (O), hydrogen (H) and sulphur (S) (5 specific input variables). Moreover, a multivariate linear regression (MLR) and a multilayer perceptron-type (MLP) artificial neural network (ANN) were adjusted to the observed data as well as known empirical correlations for comparison purposes. The current investigation has produced two results. The MARS model is used to first demonstrate the significance (or strength) of each input variable on the coal HHV (output variable). Second, the most accurate predictor of the coal HHV was the MARS-relied approximation. In fact, using this method on coal testing samples resulted in a MARS regression with coefficients of determination and correlation coefficients for the coal HHV estimation of 0.9921 and 0.9960, respectively. The agreement between the data that were observed and those that were predicted using the GS/MARS-relied approximation proved that the latter had performed satisfactorily.