As a principal energy globally, coal’s quality
and variety
critically influence the effectiveness of industrial processes. Different
coal types cater to specific industrial requirements due to their
unique attributes. Traditional methods for coal classification, typically
relying on manual examination and chemical assays, lack efficiency
and fail to offer consistent accuracy. Addressing these challenges,
this work introduces an algorithm based on the reflectance spectrum
of coal and machine learning. This method approach facilitates the
rapid and accurate classification of coal types through the analysis
of coal spectral data. First, the reflection spectra of three types
of coal, namely, bituminous coal, anthracite, and lignite, were collected
and preprocessed. Second, a model utilizing two hidden layer extreme
learning machine (TELM) and affine transformation function is introduced,
which is called affine transformation function TELM (AT-TELM). AT-TELM
introduces an affine transformation function on the basis of TELM,
so that the hidden layer output satisfies the maximum entropy principle
and improves the recognition performance of the model. Third, we improve
AT-TELM by optimizing the weight matrix and bias of AT-TELM to address
the issue of highly skewed distribution caused by randomly assigned
weights and biases. The method is named the improved affine transformation
function (IAT-TELM). The experimental findings demonstrate that IAT-TELM
achieves a remarkable coal classification accuracy of 97.8%, offering
a cost-effective, rapid, and precise method for coal classification.