Berkeley Pit Data Analysis Using Neural Networks
Principal Investigator: Dr. Curtis Link
clink@mtech.edu
This project used artificial neural network (ANN) for analysis of
geochemical and similar data sets such as those acquired from the
Berkeley Pit, Butte, MT. There are two main types of ANNs and both
lend themselves to analyses of this nature. These two types are
described as 1) supervised and 2) unsupervised networks. Supervised
networks are used in conjunction with or in place of conventional
prediction models. They require sets of known inputs and target
results or measurements. Unsupervised networks serve a useful function
as data mining tools. They do not require pairs of input/target
values but instead make an unbiased determination of groups or clusters
that occur in the data. Both supervised and unsupervised ANN were
utilized in this project. The unsupervised ANN approach was not
successful due to the limited data set available for input. The
supervised ANN which was constructed and trained to investigate
relationships between various chemical species, depth, pH, and specific
conductivity produced encouraging results. It was determined that
the supervised ANN approached should be pursued with a more complete
data set.
Activity IV, Project 14
Final Report
|
