2021 Virtual Undergraduate Research Symposium
2021 Virtual Undergraduate Research Symposium
Machine Learning and AI for Resource Engineering
Machine Learning and AI for Resource Engineering
PROJECT NUMBER: 80 | AUTHOR: Madhumitha Thiruvenkadam, Computer Science
MENTOR: Sebnem Duzgun, Mining Engineering
GRADUATE STUDENT MENTOR: Jaime Moraga, Mining Engineering
ABSTRACT
The purpose of the research was to develop machine learning and artificial intelligence multimodels to identify patterns and perform predictions in the field of geosciences and resources engineering. The multimodels were applied to the classification of land and satellite imagery on the geothermal energy potential of two sites in Nevada. Using the large data sets collected, training and testing or resampling techniques were used to estimate the error rates of the classification methods. Once the model was trained using support vector machines, it was used to predict the target variable (geothermal) using various features such as temperature, minerals, faults, subsidence, and uplift. Dimensionality reduction technique was applied, thus, only certain features were selected to fit the model. It was found that the most contributing feature in this classification was the faults. Ultimately, feature selection reduced the computational cost of the modeling and even improved the performance and the accuracy of the model.
PRESENTATION
AUTHOR BIOGRAPHY
Madhumitha Thiruvenkadam (Madhu) is a sophomore at Colorado School of Mines. She is pursuing a bachelor’s degree in Computer Science with a Data Science track. Concurrently, she is obtaining a master’s degree in the field of Data Science and Machine learning.
She has been involved with the Mines Undergraduate research fellowship program in Machine Learning and AI for Resources Engineering. She has been working on this project, which revolves around using different Machine Learning techniques on Geosciences, Mining, Energy, and/or Environmental data. In particular, applying this to the specific problem of classification of geothermal energy potential with ML and remote sensing in two sites in Nevada.
She has enjoyed the experience and would like to continue applying different Machine learning techniques such as regression, clustering, etc. In the meantime, she enjoys hiking, playing tennis, skiing, and spending time in the mountains.
I enjoyed reading your abstract and your poster! Unfortunately, I could not get your video to play on my MacBook Pro. Based on the results of this research, do you think your research could be applied to exploration for oil and natural gas? If so, then could you speculate as to which three variables might be the most relevant after dimensionality reduction is applied?
Thank you for your question! I strongly believe that my research can be applied to exploration for oil and natural gas. I believe this can be implemented using the play fairway analysis methodology. This technique can be used to reduce risk by establishing areas with a higher potential for failure. Once we know the actual results after the prediction and exploration, we can train the model with the new data using machine learning techniques such as Support Vector Machines (SVM) and estimate the weights of each factor.
Nicely done Madhumitha! As a PE student, I think this is a very interesting topic and I am glad that it can be applied to the industry. I am curious to know what kind of data could be obtain using the results of your research to PE. For example, can it be applied to reservoir characterization?
Thank you for your interest in my project! I certainly believe this can be applied to petroleum engineering. I believe that deep learning concepts and Artificial Neural Networks can be implemented in the domain of reservoir characterization. One way is to use the results of the chemical analyses and use remote sensing methods to characterize fumes or gases from the sites/wells.