A deep exploration of locational market price (LMP) data focused on Madison, WI for Madison Gas & Electric (MGE). Multiple data sources were used and coordinated with data engineers to develop a temporal fusion transformer that captures the spatiotemporal nature of LMP data, allows for a 2-week forecast, and maintains explainability. This is used by MGE energy traders to maximize potential revenue. PyTorch and Tensorflow were both used for the temporal fusion transformer, and a large quantity of SQL and BeanShell (Java child language) was written to move and download data.
Code is restricted. Ask me for details.