A Project Peek: Leveraging Machine Learning in the Energy Markets

Umar Ali

With the right data in hand, machine learning can provide powerful insights.

Machine learning is a powerful tool, but it’s only as good without proper data modeling. Without well-organized and structured records of your input variables (or “features”), the machine will be unable to make any predictions or evaluations on its own. In this project, we had to source some information from outside sources and use an API connection for energy prices across America’s states – this is what gave us a better understanding of how prices can change drastically depending on which ISO we look at.  The data modeling process begins by exploring the most volatile nodes from each ISO, looking at times when prices are high and low. This helps us find an indicator to show which direction the energy markets may go in next based on their current state.

The Machine Learning project began when we prepped and analyzed data. Once this was completed, many different models including decision trees; KNN analysis (knowledge neighbor networks); neural networking became active in creating a model that would provide us with an accurate prediction of future outcomes based on past observations.

Mining historical patterns from large sets involve identifying regularities or trends which can then be applied to new cases so they may have meaningful insights drawn out about them. Error Metrix is a great tool for gaining confidence in your pricing model. We used Mean Absolute Error to predict how prices will change based on historical data, but because energy costs can vary so much from one ISO’s average cost of $2 per kilowatt hour (kWh) we also had to take into account Root Mean Squared errors when accounting for sudden changes within the grid system as well

To create a data visualization for this project, we had to first build our model. We created the geographical heat map that would illustrate where energy grid prices were at their most optimum so it could help us decide on developing new assets in those areas of the highest value with lower cost-competitiveness than other parts of countries around them which may not have optimal conditions right now but will soon enough as technology progresses

The goal of this project was to find the most optimal location for an energy storage facility. To do so, we had three variables that needed consideration: capacity rate and congestion on electrical grids as well as node pricing- which is how much it costs per kilowatt hour (kWh). With these considerations in mind, our team created a methodical model based on data gathered from a rest API that gave us node price updates for different ISOs around North America with confidence intervals wide enough so they could be confident their solution would satisfy client needs.


Umar Ali

Umar is a Data Consultant at Pomerol Partners out of the Chicago offices. His day-to-day tasks include performing data analysis and assisting with machine learning projects.

Umar Graduated from Loyola University Chicago with a B.A in Economics, where he took part in national scaled case study competitions. He also contributed to research around the Hope IV Act and how high-rise towers can correlate to crime rates. After graduating college, he started his professional career in banking at J.P Morgan as part of the syndicated finance teams. Once moving on from working in banking, Umar decided to try consulting out. He fell in love with the field and began working as an Economic Consultant. Once he gained the experience, he needed, Umar transitioned to start his own consulting firm that focused on energy markets and forecasting. After finding out he is expecting his first child with his wife he looked for the closest opportunity where his entrepreneurial mind would be a good fit. This brings him to his current role at Pomerol Partners where he continues to grow his data analytics, and machine learning skills.