A Visual and Intuitive Microgrid Energy Management System
Our models are currently not exhibiting real world behavior closely enough and we plan to improve upon their performance in the coming weeks.
The multi-period solution trajectory is slow to converge and doesn’t show correct behavior for multiple test scenarios, therefore there is a risk that the optimizer has some deep structural issues that may not be solvable within the schedule.
System Changes
We’ve decided upon our basic visual layout for the final webapp. Which will include some sample scenarios (i.e. winy day, cloudy day, etc.) as well as a slider to travel through time. The SUGAR optimization codebase was migrated into a submodule to cleanup the overall file hierarchy.
Schedule Changes
We are mostly on schedule. Integration has taken a bit longer than expected so more work will need to be done next week to improve upon models and visuals
This week was focused on integrating the components of our project in preparation for the demo next week. For my forecasting models, this meant I had to ensure that my trained models were outputting consistent and reasonable predictions given current weather forecasts. To make this easier, i created a main file with three functions (one for each of solar, load, and wind forecasting) with each function taking in a location and returning the respective predictions in a pandas dataframe, indexed by hour. This way the web app and the optimization solver and access the forecasts from one place.
For next week, I’m looking to seriously improve the performance of my models, since the load and wind predictions seem to be especially overfitting. One idea that I’ve started experimenting with is using an autoML tool which I’ve been helping develop as part of my research at the Auton Lab. I’ve already run our wind data through this autoML system and identified some models that achieve better results than our current LSTM. A ranking of these models by test R^2 score can be found at the bottom of this post.
The team has successfully progressed for each component of the project. However, the ongoing development and integration of the side panel and ML forecasting model could introduce complexity. Additionally, the challenge of overfitting in the load data forecasting model indicates a risk in the accuracy and reliability of ML predictions.
Significant system changes include the initialization of an AWS EC2 Ubuntu instance, marking a pivotal shift to a more robust and scalable deployment environment. On the ML front, the refactoring and modularization of the forecasting codebase represent substantial improvements, enabling more efficient code management and model training processes.
Currently, everything aligns with the project timeline. Our team plans to begin component integration next week and try our best to meet our deadlines.
This week I focused on properly structuring my codebase for ML forecasting and developing a pipeline for our three training datasets (load, solar, wind) centered around a simple lstm model. Code refactoring involved modularizing my data preprocessing, model training, predicting, and writing outputs to files. This made it much easier to share functions across python scripts so I could easily extend my LSTM training code.
The current LSTM model is performing well for the wind and solar data NRMSE of 11.3% and 11.0% respectively. The results for the load data, however, are not yet good enough and it appears to be resulting from overfitting, which I plan to fix this weekend.
Overall I’m making good progress, but I’m a bit behind considering I planned to begin integrating my models with the optimization and web app this week. However, we plan to put together our components beginning next week and I expect to have all my models outputting relatively accurate predictions in time for the demo.
Risks
System Changes
Schedule Changes